0% found this document useful (0 votes)

7 views265 pages

AI Innovation in Medical Imaging Diagnostics

The document is a book titled 'AI Innovation in Medical Imaging Diagnostics' edited by Kalaivani Anbarasan, which explores the application of artificial intelligence in medical imaging diagnostics. It includes various chapters that discuss topics such as machine learning in healthcare, tumor detection in MRI images, and breast cancer diagnosis using advanced techniques. The book aims to provide insights into the latest research and technological advancements in medical imaging to enhance clinical practices.

Uploaded by

Marcela Aristizábal

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

7 views265 pages

AI Innovation in Medical Imaging Diagnostics

Uploaded by

Marcela Aristizábal

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 265

AI Innovation in Medical

Imaging Diagnostics

Kalaivani Anbarasan
Department of Computer Science and Engineering, Saveetha
School of Engineering, India & Saveetha Institute of Medical
and Technical Sciences, Chennai, India

A volume in the Advances in

Medical Technologies and Clinical
Practice (AMTCP) Book Series
Published in the United States of America by
IGI Global
Medical Information Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA, USA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com

Copyright © 2021 by IGI Global. All rights reserved. No part of this publication may be
reproduced, stored or distributed in any form or by any means, electronic or mechanical, including
photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the
names of the products or companies does not indicate a claim of ownership by IGI Global of the
trademark or registered trademark.

Library of Congress Cataloging-in-Publication Data

Names: Anbarasan, Kalaivani, 1975- editor.

Title: AI innovation in medical imaging diagnostics / Kalaivani Anbarasan,
editor.
Description: Hershey, PA : Medical Information Science Reference, [2020] |
Includes bibliographical references and index. | Summary: “This book
examines the application of artificial intelligence in medical imaging
diagnostics”-- Provided by publisher.
Identifiers: LCCN 2019049195 (print) | LCCN 2019049196 (ebook) | ISBN
9781799830924 (hardcover) | ISBN 9781799830931 (ebook)
Subjects: MESH: Artificial Intelligence | Diagnostic Imaging--methods |
Deep Learning | Machine Learning
Classification: LCC RC78.7.D53 (print) | LCC RC78.7.D53 (ebook) | NLM W
26.55.A7 | DDC 616.07/54--dc23
LC record available at https://lccn.loc.gov/2019049195
LC ebook record available at https://lccn.loc.gov/2019049196

This book is published in the IGI Global book series Advances in Medical Technologies and
Clinical Practice (AMTCP) (ISSN: 2327-9354; eISSN: 2327-9370)

British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.

All work contributed to this book is new, previously-unpublished material.

The views expressed in this book are those of the authors, but not necessarily of the publisher.

For electronic access to this publication, please contact: eresources@igi-global.com.

Advances in Medical
Technologies and Clinical
Practice (AMTCP) Book
Series
ISSN:2327-9354
EISSN:2327-9370

Editor-in-Chief: Srikanta Patnaik, SOA University, India; Priti Das, S.C.B.

Medical College, India
Mission
Medical technological innovation continues to provide avenues of research for faster
and safer diagnosis and treatments for patients. Practitioners must stay up to date with
these latest advancements to provide the best care for nursing and clinical practices.
The Advances in Medical Technologies and Clinical Practice (AMTCP) Book
Series brings together the most recent research on the latest technology used in areas
of nursing informatics, clinical technology, biomedicine, diagnostic technologies,
and more. Researchers, students, and practitioners in this field will benefit from this
fundamental coverage on the use of technology in clinical practices.

Coverage
• Clinical Data Mining IGI Global is currently accepting
• Biomedical Applications manuscripts for publication within this
• Nutrition series. To submit a proposal for a volume in
• Clinical Studies this series, please contact our Acquisition
• Nursing Informatics Editors at Acquisitions@igi-global.com or
• Biomechanics visit: http://www.igi-global.com/publish/.
• Clinical Nutrition
• Clinical High-Performance Computing
• Biometrics
• Telemedicine

The Advances in Medical Technologies and Clinical Practice (AMTCP) Book Series (ISSN 2327-9354) is published
by IGI Global, 701 E. Chocolate Avenue, Hershey, PA 17033-1240, USA, www.igi-global.com. This series is composed
of titles available for purchase individually; each title is edited to be contextually exclusive from any other title within the
series. For pricing and ordering information please visit http://www.igi-global.com/book-series/advances-medical-tech-
nologies-clinical-practice/73682. Postmaster: Send all address changes to above address. © © 2021 IGI Global. All rights,
including translation in other languages reserved by the publisher. No part of this series may be reproduced or used in any
form or by any means – graphics, electronic, or mechanical, including photocopying, recording, taping, or information and
retrieval systems – without written permission from the publisher, except for non commercial, educational use, including
classroom teaching purposes. The views expressed in this series are those of the authors, but not necessarily of IGI Global.
Titles in this Series
For a list of additional titles in this series, please visit: http://www.igi-global.com/book-series

Optimizing Health Monitoring Systems With Wireless Technology

Nilmini Wickramasinghe (Swinburne University of Technology, Australia & Epworth
HealthCare, Australia)
Medical Information Science Reference • © 2021 • 338pp • H/C (ISBN: 9781522560678)
• US $265.00

Deep Learning Applications in Medical Imaging

Sanjay Saxena (International Institute of Information Technology, India) and Sudip Paul
(North-Eastern Hill University, India)
Medical Information Science Reference • © 2021 • 274pp • H/C (ISBN: 9781799850717)
• US $245.00

Design and Quality Considerations for Developing Mobile Apps for Medication Management
Emerging Research and Opportunities
Kevin Yap (La Trobe University, Australia) Eskinder Eshetu Ali (Addis Ababa University,
Ethiopia) and Lita Chew (National University of Singapore, Singapore)
Medical Information Science Reference • © 2021 • 256pp • H/C (ISBN: 9781799838326)
• US $225.00

Expert Approaches to Health IT Tools in Clinical Practice

Ramgopal Kashyap (Amity University, Raipur, India)
Medical Information Science Reference • © 2021 • 300pp • H/C (ISBN: 9781799840510)
• US $245.00

Opportunities and Challenges in Digital Healthcare Innovation

Kamaljeet Sandhu (University of New England, Australia)
Medical Information Science Reference • © 2020 • 261pp • H/C (ISBN: 9781799832744)
• US $285.00

701 East Chocolate Avenue, Hershey, PA 17033, USA

Tel: 717-533-8845 x100 • Fax: 717-533-8661
E-Mail: cust@igi-global.com • www.igi-global.com
Table of Contents

Preface.................................................................................................................. xv

Chapter 1
Detection of Ocular Pathologies From Iris Images Using Blind De-
Convolution and Fuzzy C-Means Clustering: Detection of Ocular Pathologies....1
Sujatha Kesavan, Dr. M. G. R. Educational and Research Institute, India
Kanya N., Dr. M. G. R. Educational and Research Institute, India
Rajeswary Hari, Dr. M. G. R. Educational and Research Institute, India
Karthikeyan V., Dr. M. G. R. Educational and Research Institute, India
Shobarani R., Dr. M. G. R. Educational and Research Institute, India

Chapter 2
Machine Learning in Healthcare...........................................................................37
Debasree Mitra, JIS College of Engineering, India
Apurba Paul, JIS College of Engineering, India
Sumanta Chatterjee, JIS College of Engineering, India

Chapter 3
Detection of Tumor From Brain MRI Images Using Supervised and
Unsupervised Methods.........................................................................................61
Kannan S., Saveetha School of Engineering, India & Saveetha Institute
of Medical and Technical Sciences, Chennai, India
Anusuya S., Saveetha School of Engineering, India & Saveetha Institute
of Medical and Technical Sciences, Chennai, India

Chapter 4
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis, Haralick
Descriptors, and Autoencoder...............................................................................76
Maira Araujo de Santana, Universidade Federal de Pernambuco, Brazil
Jessiane Mônica Silva Pereira, Universidade de Pernambuco, Brazil
Washington Wagner Azevedo da Silva, Universidade Federal de
Pernambuco, Brazil
Wellington Pinheiro dos Santos, Universidade Federal de Pernambuco,
Brazil

Chapter 5
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis
From Lung CT Images..........................................................................................92
Beaulah Jeyavathana Rajendran, Saveetha School of Engineering, India &
Saveetha Institute of Medical and Technical Sciences, Chennai, India
Kanimozhi K. V., Saveetha School of Engineering, India & Saveetha
Institute of Medical and Technical Sciences, Chennai, India

Chapter 6
An Ensemble Feature Subset Selection for Women Breast Cancer
Classification.......................................................................................................101
A. Kalaivani, Saveetha School of Engineering, India & Saveetha
Institute of Medical and Technical Sciences, Chennai, India

Chapter 7
A Content-Based Approach to Medical Image Retrieval....................................114
Anitha K., Saveetha School of Engineering, India & Saveetha Institute of
Medical and Technical Sciences, Chennai, India
Naresh K., VIT University, India
Rukmani Devi D., RMD Engineering College, India

Chapter 8
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images
Using Intracellular Markers and Colocalization Operation................................137
Balanagireddy G., Rajiv Gandhi University of Knowledge Technologies,
India & Dr. A. P. J. Abdul Kalam Technical University, Ongole,
India
Ananthajothi K., Misrimal Navajee Munoth Jain Engineering College, India
Ganesh Babu T. R., Muthayammal Engineering College, India
Sudha V., Sona College of Technology, India

Chapter 9
Enchodroma Tumor Detection From MRI Images Using SVM Classifier.........155
G. Durgadevi, New Prince Shri Bhavani College of Engineering and
Technology, India
K. Sujatha, Dr. M. G. R. Educational and Research Institute, India
K.S. Thivya, Dr. M.G.R. Educational and Research Institute, India
S. Elakkiya, Dr. M.G.R. Educational and Research Institute, India
M. Anand, Dr. M.G.R. Educational and Research Institute, India
S. Shobana, New Prince Shri Bhavani College of Engineering and

Technology, India

Chapter 10
An Approach to Cloud Computing for Medical Image Analysis........................164
M. P. Chitra, Panimalar Institute of Technology, India
R. S. Ponmagal, SRM Institute of Science and Technology, India
N. P. G. Bhavani, Meenakshi College of Engineering, India
V. Srividhya, Meenakshi College of Engineering, India

Chapter 11
Segmentation of Spine Tumour Using K-Means and Active Contour and
Feature Extraction Using GLCM........................................................................194
Malathi M., Rajalakshmi Institute of Technology, India
Sujatha Kesavan, Dr. M. G. R. Educational Research Institute of
Technology, India
Praveen K., Chennai Institute of Technology, India

Chapter 12
A Survey on Early Detection of Women’s Breast Cancer Using IoT..................208
P. Malathi, Saveetha School of Engineering, India & Saveetha Institute
of Medical and Technical Sciences, Chennai, India
A. Kalaivani, Saveetha School of Engineering, India & Saveetha
Institute of Medical and Technical Sciences, Chennai, India

Compilation of References............................................................................... 220

About the Contributors.................................................................................... 240

Index................................................................................................................... 247
Detailed Table of Contents

Preface.................................................................................................................. xv

The images of disease-affected and normal eyes collected from high-resolution

fundus (HRF) image database are analyzed, and the influence of ocular diseases on
iris using a reliable fuzzy recognition scheme is proposed. Nearly 45 samples of
iris images are acquired using Canon CR-1 fundus camera with a field of view of
45° when subjected to routine ophthalmology visits, and the samples of eye images
include healthy eyes, eyes affected by glaucoma, cataract, and diabetic retinopathy.
These images are then subjected to various image processing techniques like pre-
processing for de-noising using blind de-convolution, wavelet-based feature extraction,
principal component analysis (PCA) for dimension reductionality, followed by fuzzy
c-means clustering inference scheme to categorize the normal and diseased eyes.
It is inferred that the proposed method takes only two minutes with an accuracy,
specificity, and sensitivity varying in the range of 94% to 98%, respectively.

Machine learning is a popular approach in the field of healthcare. Healthcare is an

important industry that provides service to millions of people and as well as at the
same time becoming top revenue earners in many countries. Machine learning in

healthcare helps to analyze thousands of different data points and suggest outcomes,
provide timely risk factors, optimize resource allocation. Machine learning is playing
a critical role in patient care, billing processing to set the target to marketing and
sales team, and medical records for patient monitoring and readmission, etc. Machine
learning is allowing healthcare specialists to develop alternate staffing models,
intellectual property management, and using the most effective way to capitalize on
developed intellectual property assets. Machine learning approaches provide smart
healthcare and reduce administrative and supply costs. Today healthcare industry
is committed to deliver quality, value, and satisfactory outcomes.

Brain tumor discovery and its segmentation from the magnetic resonance images
(MRI) is a difficult task that has convoluted structures that make it hard to section
the tumor with MR cerebrum images, different tissues, white issue, gray issue, and
cerebrospinal liquid. A mechanized grouping for brain tumor location and division
helps the patients for legitimate treatment. Additionally, the method improves the
analysis and decreases the indicative time. In the separation of cerebrum tumor, MRI
images would focus on the size, shape, area, and surface of MRI images. In this
chapter, the authors have focused various supervised and unsupervised clustering
techniques for identifying brain tumor and separating it using convolutional neural
network (CNN), k-means clustering, fuzzy c-means grouping, and so on.

In this chapter, the authors used autoencoder in data preprocessing step in an attempt
to improve image representation, consequently increasing classification performance.
The authors applied autoencoder to the task of breast lesion classification in
mammographic images. Image Retrieval in Medical Applications (IRMA) database

was used. This database has a total of 2,796 ROI (regions of interest) images from
mammograms. The images are from patients in one of the three conditions: with a
benign lesion, a malignant lesion, or presenting healthy breast. In this study, images
were from mostly fatty breasts and authors assessed different intelligent algorithms
performance in grouping the images in their respective diagnosis.

Tuberculosis is one of the hazardous infectious diseases that can be categorized by the
evolution of tubercles in the tissues. This disease mainly affects the lungs and also the
other parts of the body. The disease can be easily diagnosed by the radiologists. The
main objective of this chapter is to get best solution selected by means of modified
particle swarm optimization is regarded as optimal feature descriptor. Five stages
are being used to detect tuberculosis disease. They are pre-processing an image,
segmenting the lungs and extracting the feature, feature selection and classification.
These stages that are used in medical image processing to identify the tuberculosis.
In the feature extraction, the GLCM approach is used to extract the features and
from the extracted feature sets the optimal features are selected by random forest.
Finally, support vector machine classifier method is used for image classification.
The experimentation is done, and intermediate results are obtained. The proposed
system accuracy results are better than the existing method in classification.

Breast cancer leads to fatal diseases both in India and America and takes the lives of
thousands of women in the world every year. The patients can be easily treated if the
signs and symptoms are identified at the early stages. But the symptoms identified
at the final stage spreads in the human body, and most of the time, the cancer is
identified at the final stage. Breast cancer detected at the early stage is treated easily
rather than at the advanced stage. Computer-aided diagnosis came into existence
from 2000 with high expectations to improve true positive diagnosis and reduce false
positive marks. Artificial intelligence revolved in computing drives the attention
of deep learning for an automated breast cancer detection and diagnosis in digital

mammography. The chapter focuses on automatic feature selection algorithm for

diagnosis of women breast cancer from digital mammographic images achieved
through multi-layer perceptron techniques.

Medical images stored in distributed and centralized servers are referred to for
knowledge, teaching, information, and diagnosis. Content-based image retrieval
(CBIR) is used to locate images in vast databases. Images are indexed and retrieved
with a set of features. The CBIR model on receipt of query extracts same set of
features of query, matches with indexed features index, and retrieves similar images
from database. Thus, the system performance mainly depends on the features
adopted for indexing. Features selected must require lesser storage, retrieval time,
cost of retrieval model, and must support different classifier algorithms. Feature
set adopted should support to improve the performance of the system. The chapter
briefs on the strength of local binary patterns (LBP) and its variants for indexing
medical images. Efficacy of the LBP is verified using medical images from OASIS.
The results presented in the chapter are obtained by direct method without the aid
of any classification techniques like SVM, neural networks, etc. The results prove
good prospects of LBP and its variants.

This chapter contributes to the study of uncertainty of signal dimensions within a

microscopic image of blood sample. Appropriate colocalization indicator classifies
the leukocytes in the region of interest having ragged boundaries. Signal transduction
has been interpreted using correlation function determined fluorescence intensity in
proposed work using just another colocalization plugin (JaCoP). Dependence between
the channels in the colocalization region is being analysed in a linear fashion using

Pearson correlation coefficient. Manders split, which gives intensity, is represented

in a channel by co-localizing pixels. Overlap coefficients are also being analysed
to analyse coefficient of each channel. Li’s intensity correlation coefficient is being
used in specific cases to interpret the impact of staining.

Magnetic resonance imaging is a standard modality used in medicine for bone

diagnosis and treatment. It offers the advantage to be a non-invasive technique that
enables the analysis of bone tissues. The early detection of tumor in the bone leads on
saving the patients’ life through proper care. The accurate detection of tumor in the
MRI scans are very easy to perform. Furthermore, the tumor detection in an image
is useful not only for medical experts, but also for other purposes like segmentation
and 3D reconstruction. The manual delineation and visual inspection will be limited
to avoid time consumption by medical doctors. The bone tumor tissue detection
allows localizing a mass of abnormal cells in a slice of magnetic resonance (MR).

Cloud computing has become popular among users in organizations and companies.
Security and efficiency are the two major problems facing cloud service providers and
their customers. Cloud data allocation facilities that allow groups of users to work
together to access the shared data are the most standard and effective working styles
in the enterprises. So, in spite of having advantages of scalability and flexibility, cloud
storage service comes with confidential and security concerns. A direct method to
defend the user data is to encrypt the data stored at the cloud. In this research work,
a secure cloud model (SCM) that contains user authentication and data scheduling
approach is scheduled. An innovative digital signature with chaotic secure hashing
(DS-CS) is used for user authentication, followed by an enhanced work scheduling

based on improved genetic algorithm to reduce the execution cost.

MRI imaging technique is used to detect spine tumours. After getting the spine
image through MRI scans calculation of area, size, and position of the spine tumour
are important to give treatment for the patient. The earlier the tumour portion of the
spine is detected using manual labeling. This is a challenging task for the radiologist,
and also it is a time-consuming process. Manual labeling of the tumour is a tiring,
tedious process for the radiologist. Accurate detection of tumour is important for
the doctor because by knowing the position and the stage of the tumour, the doctor
can decide the type of treatment for the patient. Next, important consideration in
the detection of a tumour is earlier diagnosis of a tumour; this will improve the
lifetime of the patient. Hence, a method which helps to segment the tumour region
automatically is proposed. Most of the research work uses clustering techniques
for segmentation. The research work used k-means clustering and active contour
segmentation to find the tumour portion.

The internet of things is probably one of the most challenging and disruptive
concepts raised in recent years. Recent development in innovation and availability
have prompted the rise of internet of things (IoT). IoT technology is used in a wide
scope of certified application circumstances. Internet of things has witnessed the
transition in life for the last few years which provides a way to analyze both the real-
time data and past data by the emerging role. The current state-of-the-art method does
not effectively diagnose breast cancer in the early stages. Thus, the early detection
of breast cancer poses a great challenge for medical experts and researchers. This
chapter alleviates this by developing a novel software to detect breast cancer at a

much earlier stage than traditional methods or self-examination.

Compilation of References............................................................................... 220

About the Contributors.................................................................................... 240

Index................................................................................................................... 247
xv

Preface

Recent advancements in the technology of medical imaging, such as CT and MRI

scanners, are making it possible to create more detailed 3D and 4D images. These
powerful images require vast amounts of digital data to help with the diagnosis of
the patient. Artificial intelligence (AI) must play a vital role in supporting with the
analysis of this medical imaging data, but it will only be viable as long as healthcare
professionals and AI interact to embrace deep thinking platforms such as automation
in the identification of diseases in patients.
AI Innovation in Medical Imaging Diagnostics is an essential reference source
that examines AI applications in medical imaging that can transform hospitals to
become more efficient in the management of patient treatment plans through the
production of faster imaging and the reduction of radiation dosages through the
PET and SPECT imaging modalities. The book also explores how data clusters
from these images can be translated into small data packages that can be accessed
by healthcare departments to give a real-time insight into patient care and required
interventions. Featuring research on topics such as assistive healthcare, cancer
detection, and machine learning, this book is ideally designed for healthcare
administrators, radiologists, data analysts, computer science professionals, medical
imaging specialists, diagnosticians, medical professionals, researchers, and students.
Chapter 1 detects Ocular Pathologies from Iris Images using two algorithms and
their results are also given in a graphical representation. The two algorithms used
for the research study are Blind De-convolution and Fuzzy-C means Clustering.
Chapter 2 deals with the basics of machine learning techniques and different
algorithm and their application in Healthcare system. The top applications of machine
learning in healthcare are specified and ways to assess the growth of the healthcare
industry in 2019 and beyond.
Chapter 3 reports about the various supervised and unsupervised techniques for
brain tumor detection and segmentation such as K-nearest neighbor (K-NN), K-means
clustering, and also morphological operator and also specified the experimental results.
Preface

Chapter 4 performs the Breast Cancer Diagnosis in Mammograms Using Wavelet

Analysis, Haralick Descriptors and Auto encoder. The proposed work produced
improved accuracy with comparison to the state-of-art techniques. Chapter 5 covers
the Feature Selection for classification of Lung CT Images using Random Forest
Algorithm. The proposed work produced better results than the existing techniques.
Chapter 6 focused on automatic feature selection algorithm for diagnosis of women
breast cancer from digital mammographic images and breast cancer classification
are achieved through multi-layer perceptron techniques. The outcome of the paper
reduced false positive rate and improved diagnosis accuracy at a greater extent.
Chapter 7 deals with retrieval of medical images based on Content Based Approach
to Medical Image Retrieval. In Chapter 8, correlation and analysis of overlapping
leukocytes in blood cell images using intracellular markers and colocalization
operation are discussed with their experimental results.
Chapter 9 covers the enchodroma Tumor Detection from MRI Images using
SVM Classifier.
Medical Images captured through different image modality are to be stored in a
cloud storage for the researchers to utilize the images for their research work. So,
the approach to Cloud Computing for Medical Image Analysis well discussed in
Chapter 10.
In Chapter 11 discuss the segmentation Of Spine Tumour using K-Means and
Active Contour and also explains how the feature extraction can be done using Gray
Level Co-occurrence Matrix.
The emphasis of Chapter 12 is to give a solution of low cost, accurate, automated,
portable cancer screening tool that can be operated by a simple clinician. Unlike
mammography, our imaging model can be radiation free, non-touch, not painful
and works for women of all ages.
We are very much grateful to the authors and reviewers for their excellent
contributions for making this book possible. Our special thanks to IGI Global, Lindsay
Wertman and support teams especially for their excellent collaborations. I pledge
my gratitude to my college honorable administrators Director, Mrs. Ramya Deepak
and Director Academics Dr. V. Deepak for their fullest support and cooperation for
the completion of this proposal. I also take this opportunity to thank my Principal,
Dr. Ramesh and Dr. S P. Chokkalingam, Program Director for their kind support
and motivation throughout the proposal. I also extend my heartfelt thanks to Dr. K.
Sujatha, Professor from Dr. MGR University who helped me throughout this journey.
This edited book covers the automatic computer aided diagnosis of spine tumor,
women breast cancer, brain tumor, Ocular Pathologies from Iris Images. .Being an
interdisciplinary book, we hope it will be useful to a wide variety of readers and
will provide useful information to professors, researchers and students.

xvi
1

Chapter 1
Detection of Ocular
Pathologies From Iris
Images Using Blind De-
Convolution and Fuzzy
C-Means Clustering:
Detection of Ocular Pathologies

Sujatha Kesavan Rajeswary Hari

Dr. M. G. R. Educational and Research Dr. M. G. R. Educational and Research
Institute, India Institute, India

Kanya N. Karthikeyan V.
Dr. M. G. R. Educational and Research Dr. M. G. R. Educational and Research
Institute, India Institute, India

Shobarani R.
Dr. M. G. R. Educational and Research Institute, India

ABSTRACT
The images of disease-affected and normal eyes collected from high-resolution
fundus (HRF) image database are analyzed, and the influence of ocular diseases on
iris using a reliable fuzzy recognition scheme is proposed. Nearly 45 samples of iris
images are acquired using Canon CR-1 fundus camera with a field of view of 45°
when subjected to routine ophthalmology visits, and the samples of eye images include
healthy eyes, eyes affected by glaucoma, cataract, and diabetic retinopathy. These

DOI: 10.4018/978-1-7998-3092-4.ch001

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

images are then subjected to various image processing techniques like pre-processing
for de-noising using blind de-convolution, wavelet-based feature extraction, principal
component analysis (PCA) for dimension reductionality, followed by fuzzy c-means
clustering inference scheme to categorize the normal and diseased eyes. It is inferred
that the proposed method takes only two minutes with an accuracy, specificity, and
sensitivity varying in the range of 94% to 98%, respectively.

INTRODUCTION

The most accurate method for biometric authentication is iris recognition and is
most impressive worldwide, which results in creation of the distinctive identification
numbers for the people in India using ADHAAR (Dhooge & de Laey, 1989), or
Canadian border control system CANPASS (Roizenblatt et al., 2004). Like any
other organ in the human body, the eyes and iris may suffer from various diseases
like cataract, acute glaucoma, posterior and anterior synechiae, retinal detachment,
rubeosis iridis, corneal vascularization, corneal grafting, iris damage and atrophy
and corneal ulcers, haze or opacities. The eye pathologies are separated into five
groups based on the impact on iris recognition: 1) healthy without impact), 2) illness
detected but still clear and unaffected iris unaffected 3) geometric distortion 4)
distortion in iris tissue and 5) obstruction in iris tissue (Aslam et al., 2009; Borgen
et al., 2009; Dhir et al., 2010; ISO/IEC 19794-6:2011, 2011; Monro et al., 2009;
Rajendra Acharya, 2011; Yuan et al., 2007).
MIRLIN, VeriEye and OSIRIS are the three methods used for iris recognition
which is used to find the difference in the average value of the comparison scores
inferred between the healthy and disease affected eyes. The comparison scores
generated for the disease infected eyes as compared with healthy eyes is not within
the tolerable limit when these conventional schemes are used. Variation in the
comparison score may mislead in false non-match rate (Budai et al., 2013; McConnon
et al., 2012; Neurotechnology, 2012; Odstrcilik et al., 2013; Seyeddain et al., 2014;
Smart Sensors Ltd, 2013).
The various ocular diseases were detected using the database. The symptoms
and the effects of various ophthalmic disorders are discussed here. Cataract is the
common ophthalmic disorder indentified worldwide. The effect of this disease
includes blurring of the eye lens causing reduced vision, Figure 1A. This eye disease
occurs due to thickening of cornea which prevents the light from entering the lens
thereby inhibiting the vision (Aggarwal & Khare, 2015; Canadian Border Services
Agency, 2015; Haleem et al., 2015; Sutra et al., 2013; Trokielewicz et al., 2014;
Unique Identification Authority of India, n.d.).

2
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 1. Eye Disorders

(Courtesy Department of Ophthalmology of the Medical University of Warsaw).

The second kind of eye disorder is acute glaucoma which causes reduction in
the space between iris and cornea closing the boundary of the iris on the outer side,
hindering the flow of aqueous humor through the trabecular mesh work leading to
drastic increase in ocular pressure resulting in loss of vision as in Figure 1B. The
third kind is called as Posterior and anterior synechiae which occurs when the iris
is partly attached to the lens or to the cornea. This changes the shape of the pupil,
with deviation in circular shape, as in Figure 1C and 1D.
Diabetic retinopathy in Figure 2(a), results due to insulin disorders causing
diabetes. The blood vessels in the light sensitive region retina are affected. It is
because of insufficient supply of oxygen leading to blindness. If this eye disorder
is diagnosed at early stage proper treatment can be given preventing blindness. The
two major types of retinopathy are non-proliferative and proliferative retinopathy.
The less severe type is non-proliferative retinopathy which causes hemorrhage in
the retina. This produces a leak in blood serum making the retina wet which leads to
diminished vision. The severe type is Proliferative retinopathy which produces new
fragile blood vessels on the retina. These vessels frequently bleed into the vitreous,
the clear jelly in the center of the eyes causing visual problems. It is treated by laser
surgery which will reduce the progression of diabetic retinopathy and at times will
reverse visual loss causing permanent damage. If Diabetic retinopathy is identified
at early stages a better control of blood sugar can be maintained by ensuring lifestyle
modification, including abrupt weight loss, dietary changes and simple exercises
(Fuadah, Setiawan, Mengko et al, 2015; Panse et al., 2015; Sachdeva & Singh,
2015; Veras, 2015).
The painless clouding of internal lens of the eye is called as cataract which is
shown in Figure 2(b). They block the light from entering the lens, causing blindness
over time. Cataracts worsen with time leading to increase in thickness of cornea.
Light rays enter the eye through pupil and lens. The function of the lens is to focus
the light onto the retina, transmitting the visual signals to the brain through the optic
nerve. Clouding of the lens reduces the vision causing blurring of the images at any
distance. The patients describe their vision to be foggy, cloudy, or filmy. Intensity

3
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

of cataracts increases with time and only less light reaches the retina. People with
cataracts have difficulty in night driving. They are characterized by double vision
and second sight. In this situation, the cataract acts as a stronger lens, temporarily
improving the ability to see things at a close distance. Formerly the people who
needed reading glasses may no longer need them and also require frequent changes
in spectacles as the vision blurring increases with time. Surgery is the only remedy to
remove cataracts which is performed for only one eye at a time and may be required
if the related vision loss cannot be corrected with glasses or contact lenses. This
involves natural altering the cloudy lens with artificial lens. The operation is usually
safe and effective (Salam et al., 2015).

Figure 2a. Diabetic Retinopathy

Figure 2b. Cataract

4
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 2c. Glaucoma

Glaucoma (shown in Figure 2c) is a condition where the vision is lost because
the optic nerve gets affected due to increase in intraocular pressure (IOP). The two
kinds of glaucoma are open-angle and closed angle glaucoma, for long term. African-
American elderly people and those who have blood relatives suffer from this condition.
Glaucoma does not produce symptoms in the early stages and when affected, the
patients notice changes in vision. Timely treatment may inhibit further vision loss
but it cannot revert existing vision loss. Glaucoma is treated with prescription eye
drops. Occasionally, laser and surgical procedures may be employed. Early diagnosis
and treatment can help preserve vision in people affected by glaucoma (Fuadah,
Setiawan, Mengko et al, 2015).

LITERATURE SURVEY

Glaucoma is a condition where degeneration of optic nerve fibre takes place leading
to decrease in FoV. Due to pressure created in blood vessels, blood and other fluids
will be observed in eye, giving the retina an abnormal appearance. This eye disorder
is called as Diabetic retinopathy, which may result in damaged blood vessels. Cataract
is a clouding of the lens of the eye and occurs frequently in older age groups. An
ophthalmologist needs a slit camera lens in the diagnosis of cataract, which may
not be possible in rural areas. Hence the health of the sensory vision is provided
by the processing the retinal images. To detect the presence of eye diseases many
attempts are taken to extract useful information. They are summarized as follows

5
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Diagnosis of Glaucoma Using Automated

Texture and Spectra Features Extraction

To detect the presence of glaucoma at early stages, (Yuan et al., 2007) the features
like higher order spectra (HOS) and texture descriptors are extracted. These
extracted features are given to Support Vector Machine (SVM), Sequential Minimal
Optimization (SMO), random forest, and Naive Bayesian (NB) for classification. The
advantages include 91% accuracy by five-fold cross validation for images captured
using Fundus imaging equipment. The demerit is that only glaucoma is identified.

Glaucoma Classification Using Regional

Wavelet Features (RWF)

In this paper (Borgen et al., 2009) eye images are classified as normal and the one
affected by glaucoma using Regional Wavelet Features of the Optic Nerve Head
(ONH). Instead of global features RWF is more accurate within an accuracy of 93%.

Angle-Closure Glaucoma Feature

Detection for Cross Examination

Redundant features are eliminated here (Aslam et al., 2009). For supervised feature
selection Minimum Redundancy Maximum Relevance (MRMR) and unsupervised
methods using Laplacian Score (Lscore) are used for cross- examination. For
classifying Adaboost machine learning classifier is used. The main drawback is
that only small data set is used.

Automatic Localization and Contour Detection of Optical Disc

Retinal diseases like Diabetic Retinopathy require identification of optic disc proposed
a system using KL divergence matching in order to localize optic disc (Monro et
al., 2009), followed by segmentation of main blood vessel. The advantages include
location of OD with 92% accuracy with less computation time for histogram analysis.
The drawback is that it not efficient for poorly contrasted images.

Retinal Fundus Diseases Diagnosis Using Image Mining

The glaucoma and diabetic retinopathy (Dhir et al., 2010) causes loss of vision is
detected using this technique. This method uses Discrete Cosine Transform with
k-Nearest Neighbor (k-NN) to classify the normal eyes and eyes affected by glaucoma
and diabetic retinopathy with better classification accuracy.

6
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Automatic Identification of Pathological Retinas

Using SURF Descriptor and Pattern Recognition

A speedy and robust feature extraction based algorithm was developed to detect
the pixels of interest to form visual dictionaries (ISO/IEC 19794-6:2011, 2011).
Thereafter k-means clustering algorithm is used to predict whether an eye image is
normal or disease affected. This method has two advantages; they include detection
of characteristic pixels in the image and are also robust with an accuracy of 95.25%.
The major disadvantages include presence of artifacts and loss of information in
creation of visual dictionary.

AUTOMATIC SEGMENTATION OF OPHTHALMIC

IMAGES FOR AREA CALCULATION OF OPTIC DISC

The appearance of the optic disc changes depending on the severity of Glaucoma
condition. The blood vessels present in the eyeball makes the detection difficult. Hence
the optic disc region need to be segmented to calculate its area to extract appropriate
features (Rajendra Acharya, 2011) so that, early detection of Glaucoma is possible.
The is done using an adaptive mask which has multiple sizes and resolutions. The
results can be improved using fuzzy logic for segmentation. Further a hardware
implementation will help real time application.

Cataract Classification and Grading for Fundus

Images Using Computer-Aided Healthcare System

This method analyzes the retinal fundus images. The feature extraction is done
to detect the cataract present. Based on severity, cataract is categorized as mild,
moderate and severe (Neurotechnology, 2012). Wavelet transform, sketch based
methods along with direct cosine transform are used for feature extraction. The
main advantage of this method is identification of cataract and non-cataract using
spatial features based on the severity of the cataract condition. The limitation of this
method is only cataract and its severity is identified and no other eye pathologies
are detected by this method.

Automated Diagnosis of Glaucoma for Detection of

Optic Disc and Cup from Color Retinal Images

The proposed a system (McConnon et al., 2012) uses Color Fundus Image (CFI)
to analyze retinal nerve damage and to detect glaucoma. Segmentation is done to

7
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

extract the digital CFI optic disc, cup and neuro-retinal rim. Active contour model is
used for the detecting cup. CMY color space is used to extract the color information
of the pallor region in M channel. Features like vertical Cup to Disc Ratio (CDR),
Horizontal to Vertical CDR (H-V CDR), Cup to Disc Area Ratio (CDAR), and Rim
to Disc Area Ratio (RDAR) are used for classification by Support Vector Machine
(SVM), Naive Bayes (NB) and k-NN. This method is cost effective when compared
to Optical Coherence Tomography and Heidelberg Retina Tomography. Even the
low quality images are segmented effectively by this method. The k-NN clustering
algorithm gives an accuracy of 96.22%. The limitation of this method includes
dependence on contour initialization for Geodesic active contour model.

Glaucoma Detection Using Image Processing Techniques

This provides knowledge on different techniques to detect glaucoma using retinal

images (Budai et al., 2013). Many methods are compared in tabular form based on
pre-processing techniques, classifiers and success rate of the algorithm. Optimal
algorithm is proposed to detect glaucoma at early stages.

Calculation of Red Area for Determination

for Glaucoma Disease

This method identifies the Red Area Percentage (RAP) for extracting the portions
of sclera (Odstrcilik et al., 2013). For this, iris segmentation is done using Circular
Hough Transform (CHT). This method is advantageous because it used real time
face detection to detect the vessels and redness of the sclera for patients suffering
from glaucoma. The extraction of sclera is difficult because the texture of sclera
and that of the skin is same, which makes this method difficult.

SVM Based Local Vessel Features for Optic disc localization

This paper proposes a method to detect glaucoma and diabetic retinopathy by

analyzing optic disc (Smart Sensors Ltd, 2013). It is dependent on features like
blood vessel density and its orientation extracted using SVM. This method produces
98% accurate results in the presence of noise. The limitation of this method is that
bright lesions might be detected as optic disc.

8
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Diagnosis of Diabetic Retinopathy by Feature

Extraction From the Fundus Images

This paper uses a novel parameter for optic disk detection to assist early stage Diabetic
Retinopathy and lesions in fundus image using MAHM algorithm (Seyeddain et
al., 2014). This novel parameter is based on the detection of the major vessels and
its intersection to approximate the optic disk region. Color properties are used for
further analysis which serves as an efficient framework for identification of Diabetic
Retinopathy and eye hemorrhages. Using this method only Diabetic Retinopathy
is detected.

Automatic Detection of Blood Vessel in Retinal Images

Retinal disease like hypertension of blood vessels is detected using Hessian matrix
with Gaussian kernel based convolution (Seyeddain et al., 2014). The is identified
by using eigen values of Hessian matrix after the convolving image and helps in
identification of both healthy and abnormal retinal images. The demerit of this
method is vessel segmentation done without the elimination of Optic disc.

Optimal Combination of Statistical Texture

Analysis for Mobile Cataract Detection

An early detection method for cataract is proposed here and this method uses an
ophthalmologist with a slit lamp camera on hand using android smart phone with
k-NN classifier for statistical texture analysis (Trokielewicz et al., 2014). This
method detects cataract with 97% accuracy and the patients need to have a smart
phone with them for identification at the initial stage itself. The disadvantage of
this method is that only cataract is identified and no information regarding other
ophthalmic disorders.
From the elaborate survey done on ophthalmic disorders, it is inferred that many
work focuses on detecting either the stages of cataract or diabetic Retinopathy using
spatial feature extraction, Circular Hough Transform, Discrete Cosine Transform,
Wavelet transform and Hessian matrix with Gaussian kernel based convolution and
classifiers like SVM, NB, k-NN, SMO and random forest methods. In applying
all these methods it is found that the maximum efficiency achieved is only 97%
approximately.

9
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

RESEARCH GAP

The detailed literature survey has paved way to improvise the de-noising, feature
extraction technique and identification algorithms for diagnosing ophthalmic disorders
like cataract, diabetic retinopathy and glaucoma at an early stages so that the proposed
algorithm becomes robust in nature. Also the evaluation of the proposed scheme is
done with the help of certain performance measures like sensitivity and specificity.

RESEARCH HIGHLIGHTS

The research highlight includes

• Identification of early, moderate and severe stages in patients suffering from

diabetic retinopathy, cataract, glaucoma and to find out the healthy eyes
• Blind de-convolution for blur elimination
• Optimal feature extraction using Wavelet transform
• Principal Component Analysis for feature reduction
• Design of a single robust Fuzzy C-means clustering algorithm for identification
of these eye disorders with their corresponding stages

METHODOLOGY

The research gap discussed in section 3 highlighted the need for an automatic system
for Identification of the three common eye disorders prevalent among the people.
This method helps to detect the various stages of cataract, Glaucoma and diabetic
retinopathy at early stages with mild impact on the patient without causing loss of
vision, resulting in vision loss if not detected initially. The methodology for the
proposed system focuses on detecting three major eye disorders at three different
stages of abnormality (mild, moderate and severe) apart from the normal eyes. The
block diagram for eye disease identification is depicted in Figure 3. The canon
CR-1 fundus camera (Dhooge & de Laey, 1989; Roizenblatt et al., 2004) is used
for capturing the images of the eyes with three types of eye diseases. Along with
this, some images of the eyes pertaining to normal condition is also captured. These
images are preprocessed for noise elimination using wavelet transform followed by
feature extraction (wavelet co-efficients). Then the Principal Component Analysis
(PCA) is used for decreasing the multi-dimensional feature set to 2D feature set,
so that the computational complexity reduces. This feature set is used as inputs for
Fuzzy C-means clustering algorithm for diagnosis.

10
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 3. System Architecture for detection of Eye Abnormality

MATERIALS AND METHODS

The various image processing based algorithms used for diagnosing the eye disorders
are discussed in this section. The algorithms used include Blind devolution for
noise removal, wavelet transform for feature extraction, PCA for feature reduction
and finally the fuzzy C- means clustering for identification of the cataract, diabetes
retinopathy and glaucoma.

Blind De-convolution Algorithm

The information about distortion is unknown; hence Blind De-convolution Algorithm

is used. If the images are corrupted with any blur or noise, then this algorithm restores
the image and computes the Point-Spread Function (PSF). Accelerated, damped
Richardson-Lucy algorithm is used for noise removal. The values of PSF are taken
as one of the inputs for Fuzzy C-means clustering algorithm (Salam et al., 2015).

11
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Step 1: Read the input eye Images corresponding to normal eyes, eyes affected with
cataract, diabetic retinopathy and glaucoma
Step 2: Create a blur and make the images of the eyes to be a corrupted one.
Step 3: Use under and oversized PSFs to restore the blurred eye images of various
categories
Step 4: The Restored PSF of the normal and abnormal eye images are analyzed
Step 5: Improve the restoration by using true PSF
Step 6: Restore the true PSF for the undersized, oversized and exact PSF for normal
and abnormal eyes

Discrete Wavelet Transform

The two dimensional wavelet transform analysis is a remarkable part of image

processing methods which has developed tremendous interest for the researchers
in analyzing medical images with an idea to expand the assessment in the areas
like eye disease detection. In this aspect, the wavelet transform is used for sectional
analysis of the images using Haar wavelet for representation of images (Fuadah,
Setiawan, & Mengko, 2015).
The images of the eyes pertaining to normal and abnormal (affected by cataract,
diabetic retinopathy and glaucoma) categories are well examined, de-noised and
divided into various levels by using 2D Discrete Wavelet Transform which uses
convolution principle with Haar wavelet. This image decomposition process shown
in Figure 4 and is given by Z = X * I * Y, where Z is the final matrix of wavelet
coefficients, I represents an original image, X represents a matrix of row filters and
Y is tagged for matrix of column filters (Lotankar et al., 2015).
The original image when it is decomposed it contains the ‘Approximations’ which
corresponds to the low frequency components. The second part is the ‘detailed
image’ which has three parts namely the horizontal, vertical and diagonal elements.
They correspond to high frequency components of the image. Hence the wavelet
transform serves an inspiration for utilizing high and low frequency components
present in the eye image. The wavelet transform divides the eye image in the time-
frequency domain with non uniform tiles (Kumar et al., 2015).
Due to its low computing requirements, the Haar transform has been mainly
used for image processing and pattern recognition (Niwas et al., 2016). From this
reason two dimensional signal processing is an area of efficient applications of Haar
transforms due to their wavelet-like structure. As the original image was processed
to wavelet transform analysis, shown in the Figure 5.

12
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 4. Decomposition Process at level 2 for eye image with cataract

Figure 5. Decomposition of the image affected with cataract - four level

13
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Principal Component Analysis (PCA)

The n-dimensional data extracted from the eye images (both normal and abnormal
category) is reduced to two dimensional data using Principal Component Analysis
(PCA) which consists of a data set with a correlation value. The PCA computes Eigen
values, Eigen vectors based on which the covariance matrix is generated (Naveen
Kumar et al., 2016). This is an orthogonal matrix. The dimension reductionality
takes place which corresponds to number of the columns in the resultant data. If
two values are correlated then its value is +1 or -1. The positive sign denotes that
increase in one value will also increase the other value and whereas the negative
sign denotes that increase in one value decreases the other value. If both the values
are uncorrelated, then the value is ‘0’. The correlation is computed by calculating
the covariance matrix which is in turn computed using Eigen values and Eigen
vectors as Figure 6.

Figure 6. Flowchart for PCA

Fuzzy C- Means Clustering Algorithm

1. Choose a large value of membership function and classify each feature value
into the cluster
2. Obtain the characteristic plot for clustered feature set values and cluster centers
extracted from normal and abnormal eye images.

14
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

3. The overlap of the membership functions using Fuzzy is inferred between

Clusters
4. If there is no overlap between the clusters, then the identification of eye disorder
is efficiently done

RESULTS AND DISCUSSION

The data base containing the eye images corresponding to both normal and abnormal
categories are acquired using Canon CR-1 fundus camera with a field of view of
45° (Budai et al., 2013). The abnormal categories of images include eye images
affected by cataract, diabetic retinopathy and glaucoma. The database consists of
totally 121 images of eyes pertaining to normal and abnormal conditions as shown
in Table 1. Nearly, 60 images of the eyes corresponding to healthy condition, eyes
affected by cataract, diabetic retinopathy and glaucoma are used for training and
the remaining 61 images are used for testing the proposed algorithms.

Preprocessing

These images are preprocessed for noise removal using De-convolution algorithm
for which the results are shown in Figure 7 (a) to (j) for healthy and abnormal eyes
respectively. If noise is present in the captured images, it causes blurring which is
reflected in its Power Spectral Function (PSF). The initial, undersized and oversized
PSF values for all the disease affected eye images are calculated and tabulated in Table
2 which serves as a feature for identifying the normal and abnormal eye conditions.

Feature Extraction

The noise removal is followed by feature extraction which includes features like mean,
median, maximum intensity, minimum intensity, range, standard deviation, Median
absolute deviation, mean absolute deviation, L1 norm, L2 norm and maximum norm
respectively. This is done using wavelet tool box in MATLAB.

Feature Reduction Using PCA

The extracted features are in n-dimensional space. Hence it has to be reduced to a 2D

feature set using PCA. These Eigen values which help to compute the co-variance
matrix and thereby the correlation co-efficients are used as the distinct feature set
which is clustered into groups by Fuzzy C-means algorithm.

15
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Table 1. Fundus Data base for Images of Eyes

16
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7a. Results for De-convolution for normal Eyes

17
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7b. Results for De-convolution for Eyes affected by Mild Cataract

18
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7c. Results for De-convolution for Eyes affected by moderate Cataract

19
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7d. Results for De-convolution for Eyes affected by Severe Cataract

20
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7e. Results for De-convolution for Eyes affected by mild Glaucoma

21
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7f. Results for De-convolution for Eyes affected by moderate Glaucoma

22
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7g. Results for De-convolution for Eyes affected by severe Glaucoma

23
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7h. Results for De-convolution for Eyes affected by mild Diabetic Retinopathy

24
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7i. Results for De-convolution for Eyes affected by moderate Diabetic
Retinopathy

25
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 7j. Results for De-convolution for Eyes affected by severe Diabetic Retinopathy

26
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Table 2. PSF values from De-convolution

PSF Values PSF Values for

PSF Values PSF Values for
Condition of for Eyes Eyes Affected
S. No for Healthy Eyes Affected
Abnormality Affected by by Diabetic
Eyes by Glaucoma
Cataract Retinopathy
1. 54.7257 42.9867 39.0857 27.1806
2. 54.7257 42.8554 39.6814 27.7359
3. 54.7257 Mild 42.1857 39.1085 27.8245
4. 54.7257 42.9678 39.2681 27.2679
5. 54.7257 42.4855 39.0852 27.3838
6. 54.7257 41.6887 38.2866 26.7359
7. 54.7257 41.45854 38.5142 25.2602
8. 54.7257 Moderate 41.7187 38.8215 25.8067
9. 54.7257 41.8967 38.2681 25.0611
10. 54.7257 41.8554 38.0852 25.1806
11. 54.7257 40.2478 37.1237 25.7359
12. 54.7257 40.3176 37.2278 24.1806
13. 54.7257 Severe 40.4687 37.6791 24.7359
14. 54.7257 40.5687 37.3478 24.1806
15. 54.7257 40.1672 37.5689 24.7359

Figure 8. Feature extraction using wavelet transform for normal eyes: (a). Eyes
affected by cataract at mild stage; (b). Eyes affected by cataract at moderate stage;
(c). Eyes affected by cataract at severe stage

27
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 9. Feature extraction using wavelet transform for eyes affected by cataract:
(a). Eyes affected by Diabetic Retinopathy at mild stage; (b). Eyes affected by
Diabetic Retinopathy at moderate stage; (c). Eyes affected by Diabetic Retinopathy
at severe stage

28
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 10. Feature extraction using wavelet transform for eyes affected by Diabetic
Retinopathy: (a). Eyes affected by Glaucoma at mild stage; (b). Eyes affected by
Glaucoma at moderate stage; (c). Eyes affected by Glaucoma at severe stage

29
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 11. Feature extraction using wavelet transform for eyes affected by Glaucoma

30
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Table 3. Reduced feature set values using PCA

Correlation
Correlation
Correlation Correlation Co-efficient
Co-efficient
Co-efficient Condition of co-Efficient for for Eyes
S. No for Eyes
for Healthy Abnormality Eyes Affected Affected
Affected by
Eyes by Cataract by Diabetic
Glaucoma
Retinopathy
1. 0. 3435 0. 2132 0.7085 0. 5425
2. 0. 3524 0. 2546 0.7681 0. 5425
3. 0.4218 Mild 0. 2347 0.7915 0. 6542
4. 0.3085 0. 1542 0.7268 0. 6412
5. 0.3681 0. 1542 0.7852 0. 6252
6. 0.3915 0. 2422 0.7866 0.5734
7. 0.3268 0. 2253 0.8851 0.6684
8. 0.4171 Moderate 0.2659 0.8882 0.5745
9. 0.4189 0.2526 0.9826 0.5311
10. 0.4185 0. 1257 0.9808 0.6128
11. 0.4024 0. 1128 0.9712 0.6769
12. 0.4017 0.2511 0.9722 0.6435
13. 0.4046 Severe 0.2586 0.9679 0.5473
14. 0.4056 0.2573 0.9734 0.6486
15. 0.4016 0.2418 0.9689 0.6359

FUZZY C-MEANS CLUSTERING FOR

EYE DISEASE IDENTIFICATION

Clustering using Fuzzy C-means algorithm is defined on the basis of extracted

features from four categories of eye images grouped as healthy eyes, eyes affected
by Glaucoma, Diabetic Retinopathy and Cataract. The output for the fuzzy C means
algorithm is shown in Figure 12. During the clustering process using the Eigen
values from PCA, the covariance matrix is computed from which the correlation
coefficients are computed. The clustering process is dependent on the correlation
efficients in Table 3 which indicates that for healthy eyes it is in the range of 0.3-0.4,
whereas for the disease affected eyes it approximately in the range of 0.1-0.2 for
cataract, 0.7-0.9 for diabetic retinopathy and 0.5-0.6 for eyes affected by glaucoma.
This is illustrated in Figure 12.

31
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Figure 12. Output for fuzzy C-means clustering algorithm

Evaluation of the Proposed Technique With Existing Methods

The evaluation results in Table 4, states that the sensitivity, specificity and accuracy
is in the range of 94% to 98% for the proposed method in comparison with existing
methods.

Table 4a. Evaluation Results for Eye disease Identification during training for the
proposed method

Diabetic
Performance Measure Healthy Eyes Glaucoma Cataract
Retinopathy
Sensitivity 94.61% ± 0.54% 94.64% ± 0.59% 98.02% ± 0.23% 95.82% ± 0.42%
Specificity 97.50% ± 0.25% 96.19% ± 0.37% 96.38% ± 0.39% 95.44% ± 0.42%
Accuracy 95.39% ± 0.51% 94.45% ± 0.64% 94.97% ± 0.61% 96.65% ± 0.33%

32
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Table 4b. Evaluation Results for Eye disease Identification during testing

Author Dataset Sensitivity Specificity Accuracy

70.99% ± 94.81% ±
Budai et. al. [budai09] Healthy eyes 92.45% ± 0.44%
4.01% 0.56%
93.81% ±
Budai et. al. [budai09] Diabetic retinopathy 72.29% ± 3.4% 93.74% ± 0.44%
0.67%
92.81% ±
Budai et. al. [budai09] Glaucoma 73.34% ± 2.3% 91.45% ± 0.44%
0.71%
74.99% ± 90.81% ±
Budai et. al. [budai09] Cataract 90.65% ± 0.44%
1.42% 0.86%
Odstrcilik et. al. 78.61% ± 95.39% ±
Healthy eyes 97.50% ± 0.65%
[odstrcilik09] 3.92% 0.61%
Odstrcilik et. al. 74.63% ± 94.45% ±
Diabetic retinopathy 96.19% ± 0.77%
[odstrcilik09] 5.66% 0.84%
Odstrcilik et. al. 79.00% ± 94.97% ±
Glaucoma 96.38% ± 0.69%
[odstrcilik09] 3.18% 0.61%
Odstrcilik et. al. 75.82% ± 94.65% ±
Cataract 95.44% ± 0.82%
[odstrcilik09 4.32% 0.54%
93.52% ± 94.26% ±
Proposed method Healthy eyes 96.39% ± 0.25%
0.54% 0.51%
93.44% ± 93.11% ±
Proposed method Diabetic retinopathy 95.21% ± 0.37%
0.59% 0.64%
97.13% ± 93.97% ±
Proposed method Glaucoma 95.49% ± 0.39%
0.23% 0.61%
94.73% ± 95.76% ±
Proposed method Cataract 94.32% ± 0.42%
0.42% 0.33%

CONCLUSION

In eye disease identification, an image processing model for detection of different

diseases such as glaucoma, cataract and diabetic retinopathy is proposed. A
combination of image processing and clustering algorithms are used where the
reduced feature set from PCA are extracted from the training dataset and then
applied to fuzzy C-means clustering algorithm to identify the exact eye disease or
the healthy condition. This method integrates a single approach with the help of
correlation coefficient for all three eye diseases and also to detect the normal eye
conditions. Different algorithms like De-convolution, wavelet transform, PCA and
fuzzy C-means clustering are used for preprocessing, feature extraction, reduction and
classification from the fundus database images. Glaucoma and Diabetic Retinopathy
are among the leading cause for blindness as compared to Cataract. But detection
of these diseases at earliest stage and treatment can aid patient in avoiding vision

33
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

loss. An automatic eye disease detection system can help by providing accurate and
early diagnosis.

REFERENCES

Aggarwal, M. K., & Khare, V. (2015). Automatic localization and contour detection of
Optic disc. 2015 International Conference on Signal Processing and Communication
(ICSC). 10.1109/ICSPCom.2015.7150686
Aloudat & Faezipour. (2016). Determination for Glaucoma Disease Based on Red
Area Percentage. 2016 IEEE Long Island Systems, Applications and Technology
Conference (LISAT).
Aslam, T. M., Tan, S. Z., & Dhillon, B. (2009). Iris recognition in the presence of
ocular disease. Journal of the Royal Society, Interface, 6(34), 2009. doi:10.1098/
rsif.2008.0530 PMID:19324690
Borgen, H., Bours, P., & Wolthusen, S. D. (2009). Simulating the Influences of
Aging and Ocular Disease on Biometric Recognition Performance. International
Conference on Biometrics 2009, 8(8), 857–867. 10.1007/978-3-642-01793-3_87
Budai, A., Bock, R., Maier, A., Hornegger, J., & Michelson, G. (2013). Robust Vessel
Segmentation in Fundus Images. International Journal of Biomedical Imaging.
Canadian Border Services Agency. (2015). CANPASS Air. Available: http: //www.
cbsa-asfc.gc.ca/prog/canpass/canpassair-eng.html
Dhir, L., Habib, N. E., Monro, D. M., & Rakshit, S. (2010). Effect of cataract
surgery and pupil dilation on iris pattern recognition for personal authentication. Eye
(London, England), 24(6), 1006–1010. doi:10.1038/eye.2009.275 PMID:19911017
Dhooge, M., & de Laey, J. J. (1989). The ocular ischemic syndrome. Bulletin de la
Société Belge d’Ophtalmologie, 231, 1–13. PMID:2488440
Elbalaoui, Fakir, Taifi, & Merbohua. (2016). Automatic Detection of Blood Vessel
in Retinal Images. 13th International Conference Computer Graphics, Imaging
and Visualization.
Fuadah, Setiawan, & Mengko. (2015). Mobile Cataract Detection using Optimal
Combination of Statistical Texture Analysis. 4th International Conference on
Instrumentation, Communications, Information Technology, and Biomedical
Engineering (ICICI-BME).

34
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Fuadah, Setiawan, Mengko, & Budiman. (2015). A computer aided healthcare system
for cataract classification and grading based on fundus image analysis. Elsevier
Science Publishers B. V.
Haleem, M. S., Han, L., van Hemert, J., & Fleming, A. (2015). Glaucoma Classification
using Regional Wavelet Features of the ONH and its Surroundinga. 37th Annual
International Conference of the IEEE Engineering in Medicine and Biology Society
(EMBC).
ISO/IEC 19794-6:2011. (2011). Information technology – Biometric data interchange
formats – Part 6: Iris image data.
Kumar, Manjunathand, & Sheshadri. (2015). Feature extraction from the fundus
images for the diagnosis of diabetic retinopathy. International Conference on
Emerging Research in Electronics. Computer Science and Technology.
Lotankar, M., Noronha, K., & Koti, J. (2015). Detection of Optic Disc and Cup from
Color Retinal Images for Automated Diagnosis of Glaucoma. IEEE UP Section
Conference on Electrical Computer and Electronics (UPCON).
McConnon, G., Deravi, F., Hoque, S., Sirlantzis, K., & Howells, G. (2012). Impact
of Common Ophthalmic Disorders on Iris Recognition. 2012 5th IAPR International
Conference on Biometrics Compendium, 277–282.
Monro, D. M., Rakshit, S., & Zhang, D. (2009). DCT-Based Iris Recognition.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 29(4), 586–595.
doi:10.1109/TPAMI.2007.1002 PMID:17299216
Naveen Kumar, B., Chauhan, R. P., & Dahiya, N. (2016). Detection of Glaucoma
using Image processing techniques: A Review. 2016 International Conference on
Microelectronics, Computing and Communications (MicroCom).
Neurotechnology. (2012). VeriEye SDK, v. 4.3. Available: https://www.
neurotechnology.com/verieye.html
Niwas, Lin, Kwoh, Kuo, Sng, Aquino, & Chew. (2016). Cross-examination for
Angle-Closure Glaucoma Feature Detection. IEEE Journal of Biomedical and
Health Informatics.
Odstrcilik, J., Budai, A., Kolar, R., & Hornegger, J. (2013, June). Retinal vessel
segmentation by improved matched filtering: Evaluation on a new high-resolution
fundus image database. IET Image Processing, 7(4), 373–383. doi:10.1049/iet-
ipr.2012.0455

35
Detection of Ocular Pathologies From Iris Images Using Blind De-Convolution

Panse, N. D., Ghorpade, T., & Jethani, V. (2015). Retinal Fundus Diseases Diagnosis
using Image Mining. IEEE International Conference on Computer, Communication
and Control (IC4-2015). 10.1109/IC4.2015.7375721
Rajendra Acharya, U. (2011, May). Automated Diagnosis of Glaucoma Using Texture
and Higher Order Spectra Features. IEEE Transactions on Information Technology
in Biomedicine, 15(3).
Roizenblatt, R., Schor, P., Dante, F., Roizenblatt, J., & Jr, R. B. (2004). Iris recognition
as a biometric method after cataract surgery. BioMedical Engineering Online, 3(2).
www.biomedical-engineering-online/com/content/3/1/2
Sachdeva, & Singh. (2015). Automatic Segmentation and Area Calculation of Optic
Disc in Ophthalmic Images. 2nd International Conference on Recent Advances in
Engineering & Computational Sciences (RAECS).
Salam, Akram, Abbas, & Anwar. (2015). Optic Disc Localization using Local Vessel
Based Features and Support Vector Machine. IEEE 15th International Conference
on Bioinformatics and Bioengineering (BIBE).
Seyeddain, O., Kraker, H., Redlberger, A., Dexl, A. K., Grabner, G., & Emesz, M.
(2014). Reliability of automatic biometric iris recognition after phacoemulsification
or drug-induced pupil dilation. European Journal of Ophthalmology, 24(1), 58–62.
doi:10.5301/ejo.5000343 PMID:23873488
Smart Sensors Ltd. (2013). MIRLIN SDK, 2, 23.
Sutra, G., Dorizzi, B., Garcia-Salitcetti, S., & Othman, N. (2013, April 23). A
biometric reference system for iris. OSIRIS version 4.1. Available: http://svnext.
it-sudparis.eu/svnview2-eph/ref syst/Iris Osiris v4.1/
Trokielewicz, M., Czajka, A., & Maciejewicz, P. (2014). Cataract influence
on iris recognition performance. Proc. SPIE 9290, Photonics Applications in
Astronomy, Communications, Industry, and High-Energy Physics Experiments.
doi:10.1117/12.2076040
Unique Identification Authority of India. (n.d.). AADHAAR. Available: https://uidai.
gov.in/what-is-aadhaar.html
Veras, R. (2015). SURF descriptor and pattern recognition techniques in automatic
identification of pathological retinas. 2015 Brazilian Conference on Intelligent
Systems.
Yuan, X., Zhou, H., & Shi, P. (2007). Iris recognition: A biometric method after
refractive surgery. Journal of Zhejiang University. Science A, 8(8), 1227–1231.
doi:10.1631/jzus.2007.A1227

36
37

Chapter 2
Machine Learning
in Healthcare
Debasree Mitra
https://orcid.org/0000-0003-3723-9499
JIS College of Engineering, India

Apurba Paul
JIS College of Engineering, India

Sumanta Chatterjee
JIS College of Engineering, India

ABSTRACT
Machine learning is a popular approach in the field of healthcare. Healthcare is an
important industry that provides service to millions of people and as well as at the
same time becoming top revenue earners in many countries. Machine learning in
healthcare helps to analyze thousands of different data points and suggest outcomes,
provide timely risk factors, optimize resource allocation. Machine learning is playing
a critical role in patient care, billing processing to set the target to marketing and
sales team, and medical records for patient monitoring and readmission, etc. Machine
learning is allowing healthcare specialists to develop alternate staffing models,
intellectual property management, and using the most effective way to capitalize on
developed intellectual property assets. Machine learning approaches provide smart
healthcare and reduce administrative and supply costs. Today healthcare industry
is committed to deliver quality, value, and satisfactory outcomes.

DOI: 10.4018/978-1-7998-3092-4.ch002

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Machine Learning in Healthcare

INTRODUCTION

What Is Machine Learning?

Machine learning (ML) explores algorithms that learn from data, builds models data
and that model used for prediction, decision making or solving task. A computer
program is to learn from experience E with respect to some class of task T and
performance P. There are two components in ML i.e. learning module and reasoning
module. Learner module takes input as experienced data and background knowledge
and builds model. Models are used by reasoning module and reasoning module
comes up with solution to the task and performance measure. Machine Learning
algorithms can generate a mathematical model based on experience data known as
training data to predict or decisions.

Figure 1. Traditional Programming vs Machine Learning

Machine learning algorithms are used in diagnose disease, banking system,

healthcare, email filtering, and computer vision, data mining, robot control,
Natural Language Processing, Speech Recognition, Machine Translation, Business
Intelligence, Fraud Detection, Consumer sentiment etc where it is very helpful to
develop an algorithm of specific instructions for performing the task. Machine
learning is related to statistics and probability, which focuses on making predictions
using computers.

What Is Healthcare?

Healthcare is the upgradation of health via technology for people. Health care is
delivered by health professionals in allied health fields. Physicians and physician
associates are a part of these health professionals. Dentistry, pharmacy, midwifery,

38
Machine Learning in Healthcare

nursing, medicine, optometry, audiology, psychology, occupational therapy, physical

therapy and other health professions are all part of health care. It includes work
done in providing primary care, secondary care, and tertiary care, as well as in
public health.
Access to health care may vary across countries, communities, and individuals,
largely influenced by social and economic conditions as well as health policies.
Providing health care services means “the timely use of personal health services to
achieve the best possible health outcomes” (Anthony & Bartlet, 1999). Factors to
consider in terms of healthcare access include financial limitations (such as insurance
coverage), geographic barriers (such as additional transportation costs, possibility
to take paid time off of work to use such services), and personal limitations (lack
of ability to communicate with healthcare providers, poor health literacy, low
income) (Langley, 1996). Limitations to health care services affects negatively the
use of medical services, efficacy of treatments, and overall outcome (well-being,
mortality rates).
Health care systems are organizations established to meet the health needs of
targeted populations. According to the World Health Organization (WHO), a well-
functioning health care system requires a financing mechanism, a well-trained
and adequately paid workforce, reliable information on which to base decisions
and policies, and well maintained health facilities to deliver quality medicines and
technologies (Muller & Guido, n.d.).
An efficient health care system can contribute to a significant part of a country’s
economy, development and industrialization. Health care is conventionally regarded
as an important determinant in promoting the general physical and mental health
and well-being of people around the world. An example of this was the worldwide
eradication of smallpox in 1980, declared by the WHO as the first disease in human
history to be completely eliminated by deliberate health care interventions.

Purpose of Machine Learning in Healthcare

Machine learning has virtually endless applications in the healthcare industry. Today,
machine learning is helping to streamline administrative processes in hospitals, map
and treat infectious diseases and personalize medical treatments.
The healthcare sector has long been an early adopter of and benefited greatly
from technological advances. These days, machine learning (a subset of artificial
intelligence) plays a key role in many health-related realms, including the development
of new medical procedures, the handling of patient data and records and the treatment
of chronic diseases. As computer scientist Sebastian Thrum told the New Yorker
in a recent article titled “A.I. Versus M.D., “Just as machines made human muscles

39
Machine Learning in Healthcare

a thousand times stronger, machines will make the human brain a thousand times
more powerful.”

Machine Learning Timeline in Healthcare

Artificial Intelligence is a broad scientific discipline combination with mathematics

and computer science that goals to understand and develop an automated systems that
provide the properties of human intelligence. Machine learning is a sub discipline of
Artificial Intelligence where machine algorithms learn by training phase and predicts
with test data. Machine learning uses a broader set of statistical techniques. Newer
techniques such as Deep Learning are based on models artificial neural network with
more accurate results for more complex data. Following diagram shows different
evolution phase of machine learning in healthcare.

Figure 2. Machine learning timeline

EXPLORATION OF HEALTHCARE DATA

The World Health Organization (WHO) collects and shares data on global health for
its 194-member countries under the Global Health Observatory (GHO) initiative.
Source users have options to browse for data by theme, category, indicator and by
country. The metadata section allows for learning how data is organized. These
healthcare datasets are available online or can be downloaded in CSV, HTML, Excel,
JSON, and XML formats.Apart from that there are many resources are available like
UCI,Kraggle,NCBI, Center for Disease Control (CDC) etc. Some popular datasets
are available UCI Heart Decease Data,Diabetes, Breast Cancer Data, Lymphography
Data Set,Lung Cancer,SPECT Heart Data Set,SPECTF Heart Data Set,Thyroid
Disease,Mammographic Mass, EEG Database and many more.

40
Machine Learning in Healthcare

MACHINE LEARNING TECHNIQUES IN HEALTHCARE

Figure 3. Structure of Machine Learning

Hypothesis

A hypothesis an explanation for some incidents or events. It is a provisional idea,

an educated guess that requires some evaluation. A good hypothesis is testable; it
can be either true or false. In science, a hypothesis must be falsifiable, meaning
that there exists a test whose outcome could mean that the hypothesis is not true.
The hypothesis must also be framed before the outcome of the test is known. A
good hypothesis fits the evidence and can be used to make predictions about new
observations or new situations. The hypothesis that best fits the evidence and can be
used to make predictions is called a theory, or is part of a theory. But in Hypothesis
in Science can be described as provisional explanation that fits the evidence and
can be confirmed or disproved. Statistical hypothesis tests are techniques used to
calculate a critical value called an “effect.” The critical value can then be interpreted
in order to determine how likely it is to observe the effect if a relationship does not
exist. If the likelihood is very small, then it suggests that the effect is probably real.
If the likelihood is large, then we may have observed a statistical fluctuation, and

41
Machine Learning in Healthcare

the effect is probably not real. For example, we may be interested in evaluating the
relationship between the means of two samples, e.g. whether the samples were drawn
from the same distribution or not, whether there is a difference between them. One
hypothesis is that there is no difference between the population means, based on
the data samples. This is a hypothesis of no effect and is called the null hypothesis
and we can use the statistical hypothesis test to either reject this hypothesis, or fail
to reject (retain) it. We don’t say “accept” because the outcome is probabilistic and
could still be wrong, just with a very low probability.

LEARNING ALGORITHM

Supervised Learning

Supervised learning indicates presence of a supervisor or teacher. There will be a

teacher algorithm. Supervised learning is a learning technique in which we teach
or train the machine using data which is labelled. That means, there should be an
answer regarding each question. The machine is provided with a new set of data so
that supervised learning algorithm analyses the training data and produces a correct
results from labeled data.

Unsupervised Learning

Unsupervised learning is the training of machine using information that is not labeled
and allows the algorithm to produce results on that information without guidance.
Here the functions of machine are to classify unsorted information according to
similarities, patterns and differences without any prior training of data. There is no
teacher algorithm like supervised learning algorithm.

Semi Supervised Learning

Semi-supervised learning is a class of machine learning techniques that also make

use of unlabeled data for training Here a small amount of labeled data is mixed with a
large amount of unlabeled data. Semi-supervised learning falls between unsupervised
learning (i. e. without any labeled data) and supervised learning (with training data).

42
Machine Learning in Healthcare

Reinforcement Learning Algorithm

Reinforcement learning algorithm is training from particular situation on particular

action upon it. It finds the best possible behaviour or path of various software and
machines considering a specific situation.

FINAL HYPOTHESIS

Hypothesis sets and machine learning algorithms generate a final hypothesis or

sometimes we called it is a model.

Goal

Model will be compared with predicted and actual data in test by accuracy,error
rate, precision etc. Those model will give maximum accuracy and precision that
will be selected in prediction to reach the goal.

DIFFERENT MODELS OF MACHINE LEARNING

There are three types of models in machine learning.

1. Classification model: This model trained with categorized data

2. Clustering model: This model identify patterns of some group of data
3. Regression Model: This model is to find the predictor between dependent and
independent data.

Figure 4. Machine Learning Models

43
Machine Learning in Healthcare

Validation and Evaluation

Most machine learning engineers divide their data into three portions: training data,
cross-validation data and testing data. The training data is used to make sure the
machine recognizes patterns in the data, the cross-validation data is used to ensure
better accuracy and efficiency of the algorithm used to train the machine and the
test data is used to see how well the machine can predict new answers based on its
training.

Figure 5. Training set, Test Set and validation Set

SUPERVISED LEARNING TECHNIQUES

Decision Trees

Decision tree makes regression or classification models in the form of a structure

similar to that of a tree. Decision tree breaks down a dataset into smaller and
smaller subsets while at the same time an associated decision tree is developed
in an incremental manner. The final result is a tree with only two types of nodes,
decision nodes and leaf nodes.

Naïve Bayesian Algorithm

In machine learning we are often interested in selecting the best hypothesis (H) given
data (D).In a classification problem, our hypothesis (H) may be the class to assign
for a new data instance (D).One of the easiest ways of selecting the most probable
hypothesis given the data that we have that we can use as our prior knowledge about
the problem. Bayes’ Theorem provides a way that we can calculate the probability
of a hypothesis given our prior knowledge.

Bayes’ Theorem is stated as: P(H|D) = (P(D|H) * P(H)) / P(D)

44
Machine Learning in Healthcare

Figure 6. Decision tree

Where P(H|D) is the probability of hypothesis H given the data D. This is

called the posterior probability. P(D|H) is the probability of data D given that the
hypothesis H was true.P(H) is the probability of hypothesis H being true (regardless
of the data). This is called the prior probability of H.P(D) is the probability of the
data (regardless of the hypothesis). We are interested in calculating the posterior
probability of P(H|D) from the prior probability p(H) with P(D) and P(D|H).After
calculating the posterior probability for a number of different hypotheses, you can
select the hypothesis with the highest probability. Naive Bayes can be extended to
real-valued attributes, most commonly by assuming a Gaussian distribution.This
extension of naive Bayes is called Gaussian Naive Bayes. Other functions can be used
to estimate the distribution of the data, but the Gaussian (or Normal distribution) is
the easiest to work with because you only need to estimate the mean and the standard
deviation from your training data.
In the below figure two event A and B is represented by Gauss Naïve bayes
Classification techniques by probability distribution function.

Neural Networks

Artificial Neural Networks are the most popular machine learning algorithms
nowadays. The invention of these Neural Networks took place in the 1970s.But they
have become popular due to the recent increase in computation tool like Python,
R, MATLAB etc.

45
Machine Learning in Healthcare

Figure 7. Naïve bayes Classifier

The neurons in human nervous system are able to learn from the past data and
similarly the ANN is also capable to learn from the past data or trained data and
provide responses in the form of predictions or classifications. ANNs are nonlinear
statistical models which establishes a complex relationship between the inputs
and outputs to discover a new pattern. A variety of tasks such as optical character
recognition, face recognition, speech recognition, machine translation as well as
medical diagnosis makes use of these artificial neural networks.
The basic concept is based upon three layer: Input Layers(IL),Hidden
layers(HL),Output layers(OL).The input layer receives the input information in the
form of various texts, numbers, audio files, image pixels, etc. Hidden Layers is the
middle layer where some mathematical computations are done . These hidden layers
can be single or multiple. Output Layer provides the result that we obtain through
rigorous computations performed by the middle layer.

Support Vector Machine

Support Vector Machines (SVMs) are a classification strategy. SMVs work by

transforming the training dataset into a hyperplane. There are two support vectors
which easily define classes, and their margins, which are the lines parallel to the
hyperplane defined by the shortest distance between a hyperplane and its support
vectors. SVMs are able to classify both linear and nonlinear data.The distance between

46
Machine Learning in Healthcare

the hyperplane and the nearest data point from either set is known as the margin.
The goal is to choose a hyperplane with the greatest possible margin between the
hyperplane and any point within the training set, giving a greater chance of new
data being classified correctly.

Figure 8. ANN

Figure 9. Support Vectors Machine

UNSUPERVISED LEARNING

Gaussian mixture Model

In this model we use the concept of normal distribution on subpopulation and

Gaussian distribution on overall population. Gaussian Mixture Model (GMM) does
not require the data to which the subpopulation belongs. This allows the model to
learn the subpopulations automatically. Since we do not know the assignment of
the subpopulation, it comes under unsupervised learning. With the help of GMMs,
one can extract the features from speech data, healthcare data etc track the multiple

47
Machine Learning in Healthcare

factors in cases where there are a number of mixture components and the means
that predict location of objects in a dataset. Many datasets can be easily modeled
with the help of Gaussian Distribution. Therefore, one can assume that the clusters
from different Gaussian Distributions. The core idea of model is that the data is
modeled with several mixtures of Gaussian Distributions.
The single dimension probability density function of a Gaussian Distribution
is as follows –

2
(x −µ)
1 −
y=
2
e 2σ

σ 2π

µ = Mean
σ = Standard Deviation
π = 3.14159…
e = 2.71828…

Hidden Markov Model

Hidden Markov models (HMMs) was developed by the Russian mathematician

Andrey Andreyevich Markov in the early 1970s. HMMS are based on the theory of
Bayes. HMMs are statistical models to capture hidden information from observable
sequential symbols. They have many applications in sequence analysis in genomic.
In a HMM, the system being modelled is assumed to be a Markov process with
unknown parameters, and the challenge is to determine the hidden parameters
from the observable parameters. A good HMM accurately models the real world
source of the observed real data and has the ability to simulate the source. A lot of
Machine Learning techniques are based on HMMs have been successfully applied to
problems including healthcare data, speech recognition, optical character recognition,
computational biology etc. HMMs are popular for their robust statistical foundation,
conceptual simplicity and malleability; they are adapted fit diverse classification
problems. In Computational Biology, a hidden Markov model (HMM) is a statistical
approach that is frequently used for modelling biological sequences.

Principal Components Analysis

Principal component analysis (PCA) is a statistical procedure . PCA uses an orthogonal

transformation to convert a set of observations of possibly correlated variables
into a set of values of linearly uncorrelated variables called principal components.
This transformation is defined in such a way that the first principal component

48
Machine Learning in Healthcare

has the largest possible variance and each succeeding component in turn has the
highest variance possible under the constraint that it is orthogonal to the preceding
components. The resulting vectors are an uncorrelated orthogonal basis set. PCA
is sensitive to the relative scaling of the original variables.

Figure 10. Basic Structure of Hidden Markov Model

Figure 11. PCA

49
Machine Learning in Healthcare

Figure 12. Summary about Datasets

SAMPLE CODING AND RESULTS

As a sample healthcare dataset we have consider PIMA Diabetic dataset here

import pandas as pd
from pandas.plotting import scatter_matrix
import matplotlib.pyplot as plt
from sklearn import model_selection
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier

50
Machine Learning in Healthcare

from sklearn.neighbors import KNeighborsClassifier

from sklearn.discriminant_analysis import
LinearDiscriminantAnalysis
from sklearn.naive_bayes import GaussianNB
from sklearn.svm import SVC
import seaborn as sns
url=”D:/MYPAPER/diabetescsv/diabetes.csv”
names=[‘Pregnancies’,’GlucosePlasma’,’BloodPressureDiastolic’,’
SkinThicknessTriceps’,’Insulin’,’BMI’,’DiabetesPedigreeFunction
’,’Age’,’Outcome’]
dataset=pd.read_csv(url,sep=’,’)
print(dataset.shape)
dataset.isnull()
dataset.dropna(inplace=True)
dataset.head(10)

Output

See Figure 13.

Figure 13. First 10 rows after data cleaning

RESULTS FOR DECISION TREE

x=dataset.drop(“Outcome”,axis=1)
y=dataset[“Outcome”]

51
Machine Learning in Healthcare

from sklearn import model_selection

x_train,x_test,y_train,y_test=model_selection.train_test_
split(x,y,test_size=0.30,random_state=1)
from sklearn.tree import DecisionTreeClassifier
dtmodel=DecisionTreeClassifier()
dtmodel.fit(x_train,y_train)
predictions=dtmodel.predict(x_test)
from sklearn.metrics import classification_report
print(classification_report(y_test,predictions))
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test,predictions))
Output:
precision recall f1-score support
0 0.63 1.00 0.77 146
1 0.00 0.00 0.00 85
avg / total 0.40 0.63 0.49 231
[[146 0]
[ 85 0]]

RESULTS FOR KNN CLASSIFIER

x=dataset.drop(“Outcome”,axis=1)
y=dataset[“Outcome”]
from sklearn import model_selection
x_train,x_test,y_train,y_test=model_selection.train_test_
split(x,y,test_size=0.30,random_state=1)
from sklearn.neighbors import KNeighborsClassifier
knnmodel=KNeighborsClassifier()
knnmodel.fit(x_train,y_train)
predictions=knnmodel.predict(x_test)
from sklearn.metrics import classification_report
print(classification_report(y_test,predictions))
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test,predictions))
Output:
precision recall f1-score support
0 0.80 0.88 0.84 146
1 0.75 0.62 0.68 85
avg / total 0.78 0.78 0.78 231

52
Machine Learning in Healthcare

[[128 18]
[ 32 53]]

Results for SVM Classifier

x=dataset.drop(“Outcome”,axis=1)
y=dataset[“Outcome”]
from sklearn import model_selection
x_train,x_test,y_train,y_test=model_selection.train_test_
split(x,y,test_size=0.30,random_state=1)
from sklearn.svm import SVC
svmmodel=SVC()
svmmodel.fit(x_train,y_train)
predictions=svmmodel.predict(x_test)
from sklearn.metrics import classification_report
print(classification_report(y_test,predictions))
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test,predictions))
Output:
precision recall f1-score support
0 0.63 1.00 0.77 146
1 0.00 0.00 0.00 85
avg / total 0.40 0.63 0.49 231
[[146 0]
[ 85 0]]

RESULTS FOR GAUSSIANNB CLASSIFIER

53
Machine Learning in Healthcare

print(classification_report(y_test,predictions))
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test,predictions))
Output:
precision recall f1-score support
0 0.80 0.88 0.84 146
1 0.75 0.62 0.68 85
avg / total 0.78 0.78 0.78 231
[[128 18]
[ 32 53]]

Results for LDA Classifier

x=dataset.drop(“Outcome”,axis=1)
y=dataset[“Outcome”]
from sklearn import model_selection
x_train,x_test,y_train,y_test=model_selection.train_test_
split(x,y,test_size=0.30,random_state=1)
from sklearn.discriminant_analysis import
LinearDiscriminantAnalysis
ldamodel=LinearDiscriminantAnalysis()
ldamodel.fit(x_train,y_train)
predictions=ldamodel.predict(x_test)
from sklearn.metrics import classification_report
print(classification_report(y_test,predictions))
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test,predictions))

Output

precision recall f1-score support

0 0.63 1.00 0.77 146
1 0.00 0.00 0.00 85
avg / total 0.40 0.63 0.49 231
[[146 0]
[ 85 0]]

54
Machine Learning in Healthcare

K- FOLD VALIDATION

In summary, cross-validation combines (averages) measures of fitness in prediction

to derive a more accurate estimate of model prediction performance.
In our dataset we have select 30% as validation size and rest as training dataset
splits values with 5 (5 Fold Cross validation Procedure) with a random seed value 7.

from sklearn import model_selection

from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.discriminant_analysis import
LinearDiscriminantAnalysis
from sklearn.naive_bayes import GaussianNB
from sklearn.svm import SVC
x=dataset.drop(“Outcome”,axis=1)
y=dataset[“Outcome”]
validation_size=0.30
seed = 7
x_train, x_validation, y_train, y_validation = model_selection.
train_test_split(x, y, test_size=validation_size, random_
state=seed)
seed = 7
scoring = ‘accuracy’
models = []
models.append((‘LR’, LogisticRegression()))
models.append((‘LDA’, LinearDiscriminantAnalysis()))
models.append((‘KNN’, KNeighborsClassifier()))
models.append((‘CART’, DecisionTreeClassifier()))
models.append((‘NB’, GaussianNB()))
models.append((‘SVM’, SVC()))
results = []
names = []
for name, model in models:
kfold = model_selection.KFold(n_splits=5, random_
state=seed)
cv_results = model_selection.cross_val_score(model, x_

55
Machine Learning in Healthcare

train, y_train, cv=kfold, scoring=scoring)

results.append(cv_results)
names.append(name)
msg = “%s: %f (%f)” % (name, cv_results.mean(), cv_
results.std())
print(msg)
Output:
LR: 0.776625 (0.076214)
LDA: 0.772851 (0.072747)
KNN: 0.726171 (0.062883)
CART: 0.704053 (0.084624)
NB: 0.761740 (0.064313)
SVM: 0.657547 (0.077963)

RESULT ANALYSIS

fig = plt.figure()
fig.suptitle(‘Algorithm Comparison’)
ax = fig.add_subplot(111)
plt.boxplot(results)
ax.set_xticklabels(names)
plt.show()

The above analysis shows the LDA and GNB will give the highest accuracy.
Whereas LR and KNN are also providing a reasonably good accuracy, The CART
and SVM both are not providing satisfactory accuracy. But K-Fold cross validation
algorithm is also showing nearer accuracy . Here valur of k=5 and we have calculated
accuracy as validation score and at the end we have calculated mean of accuracy.In
case of K-Fold validation for CART and SVM accuracy level have been increased
slightly and it cleared from above figure.

APPLICATIONS OF MACHINE LEARNING IN HEALTHCARE

Identifying Diseases and Medical Diagnosis and Analysis

One of the chief Machine Learning applications in healthcare is the identification

and diagnosis of diseases .This can include anything from diabetes, heart diseases,

56
Machine Learning in Healthcare

Figure 14. Algorithm Comparison

cancers which are tough to predict during the initial stages or to identify some
genetic diseases.

Medical Imaging Diagnosis

Machine learning and deep learning are both responsible for the breakthrough
technology called Computer Vision. E. g. InnerEye developed by Microsoft which
works on image diagnostic tools for image analysis. process.

Smart Health Records

Machine Learning will provide best-predicted values related to the patients in their
respected health condition and also it helps to analyze the previous health records.
For that purpose we need to maintain a repository or in other words warehouse where
we can maintain data related to the patients and their treatment. Regarding each
hospital there should be health record system. These health records of the patients
can be accessed only doctors or hospital by an identification number. This kind of
approaches required an web application with prediction system. The records can be
sequential or hierarchical. Admin can only change the order of data.

57
Machine Learning in Healthcare

Clinical Trial and Research

Machine learning has immense potential in the field of clinical trials in pharmacy
industry. It will decrease the clinical trial cost and save time. Applying ML-based
predictive analytics to identify potential clinical trial results.

Crowdsourced Data Collection

Crowdsourcing is the new trends in market. Different kind of medical data of all
the ages are collected. This live health data has great significant to researcher and
scientists now a days. IBM recently partnered with Medtronic to predict through
ML by available diabetes and insulin data in real time based on the crowdsourced
information. With the advancements of IoT and bigdata with machine learning the
healthcare industry is in booming condition. All over the world research has been
faster than previous condition.

Better Radiotherapy

One of the most widely applications of machine learning in healthcare is in the field
of Radiology. Through computer vision and medical image analysis we can model
many tissue regions, cancer foci, etc by using complex equations. Since Machine
Learning based algorithms learn from the trained dataset of different samples
available globally and hence it becomes easier to diagnose and predict the factors
responsible for cancer .As for example different classification based approaches for
prediction of cancer stages. As for example Google’s DeepMind Health is actively
helping researchers in UCLH to develop algorithms which can detect the difference
between healthy and cancerous tissue .

Outbreak Prediction

Machine learning is useful in monitoring and predicting epidemics around the

world. There are huge data collected from satellites, social media updates, website
information, etc through natural language processing. Artificial neural networks
help to gather this information and predict everything from malaria, cholera, dengue
outbreaks to severe diseases. Especially, this is helpful in third-world countries as
they lack in medical infrastructure and educational systems due to huge population.
As for example is the ProMED-mail which is an Internet based reporting system
which monitors evolving diseases and emerging ones and provides outbreak reports
in real-time.

58
Machine Learning in Healthcare

Drug Discovery and Manufacturing

Manufacturing a new drug is very expensive and a long process because they are
depended on variety of tests and their results.. With the advancements in ML machine
learning can next-generation sequencing and precision medicine can be useful to
help to cure many health diseases. Unsupervised machine learning algorithm can
identify patterns in data without providing for any predictions.

THE IDEAL SYSTEMS MACHINE LEARNING

IN HEALTHCARE: CASE STUDY

As for example a cloud-based open system such as the Health Catalyst Data Operating
System. Its aim to answer healthcares growing data needs by combining the features
of data warehousing, clinical data repositories, and HIEs in a single, common-sense
technology platform.
Following components are here

1. Cloud-Based Data Operating System

2. Health Information Exchange (HIE)
3. Hybrid Big data SQL Architecture-HTAP
4. Big Data Architecture

Figure 15. Health Catalyst Data Operating System

59
Machine Learning in Healthcare

CONCLUSION

Many sectors like finance, education, agriculture are using machine learning and
hence healthcare cannot stand behind. Google has developed an ML algorithm to
identify cancerous tutors. Stanford is using it to identify skin cancer. People should
stop thinking machine learning as a concept for future .Instead we should embrace the
tools and make the use of all opportunities . These applications of machine learning
are advancing the field of healthcare into a completely new arena of opportunities..

REFERENCES

Anthony, M., & Bartlet, P. (1999). Neural Network Learning: Theoretical Foundations.
Cambridge University Press. doi:10.1017/CBO9780511624216
Langley, P. (1996). Elements of Machine Learning. Morgan Kaufmann.
Muller, A.C., & Guido, S. (n.d.). Introduction to Machine Learning with Python.
O’Reilly.

60
61

Chapter 3
Detection of Tumor From Brain
MRI Images Using Supervised
and Unsupervised Methods
Kannan S.
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

Anusuya S.
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

ABSTRACT
Brain tumor discovery and its segmentation from the magnetic resonance images
(MRI) is a difficult task that has convoluted structures that make it hard to section
the tumor with MR cerebrum images, different tissues, white issue, gray issue, and
cerebrospinal liquid. A mechanized grouping for brain tumor location and division
helps the patients for legitimate treatment. Additionally, the method improves the
analysis and decreases the indicative time. In the separation of cerebrum tumor,
MRI images would focus on the size, shape, area, and surface of MRI images. In this
chapter, the authors have focused various supervised and unsupervised clustering
techniques for identifying brain tumor and separating it using convolutional neural
network (CNN), k-means clustering, fuzzy c-means grouping, and so on.

DOI: 10.4018/978-1-7998-3092-4.ch003

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

1. INTRODUCTION

Tumor refers to a mass of tissue which controls the development of growing of

further tissue. It is an intracranial strong neoplasm and a gathering of anomalous
cells develop inside, around the brain or cerebrum through uncontrolled cell division.
Brain is the inward piece of the focal sensory system (Aggarwal & Kaur, 2012).
Kindhearted tumor is portrayed by an ordinary shape, does not suddenly extend,
not attack nonadjacent cells and neighboring sound tissues. Moles are a case of
considerate tumors and premalignant tumor is a precancerous stage that can be viewed
as a sickness, may prompt disease if not appropriately treated. Harm is the tumor
type which develops like normal tissue that attacks solid neighboring tissues and can
eventually bring about death. The term threatening is fundamentally a restorative
term which alludes to an extreme advancing ailment and harmful tumor is utilized
to depict malignant growth. In the figure.1 shows the benign and malignant tumors.

Figure 1. Benign and Malignant Tumor

The side effects for brain tumors can be perceived by spewing, queasiness, cerebral
pain, sudden difference in character or conduct, deadness and shortcoming. Now,
loss of sensation and memory can be experienced by the patient (Aggarwal & Kaur,
2012). The brain tumor division procedure contains preprocessing, extraction of
highlights from MRI images, and division utilizing administered or solo strategies.

62
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

2. LITERATURE SURVEY

In the event that any of the evaluations, it must be recognized convenient and found
precisely (Kaur & Rani, 2016)[5]. For example, Magnetic Imaging Resonance (MRI)
and Computed Tomography (CT) are broadly used to identify the tumor. Among
these restorative imaging modalities, MRI is most broadly utilized and exceedingly
favored non-intrusive strategy in biomedical, radiology and therapeutic imaging fields
because of its ability to identify and envision better subtleties in the inner structure of
the body by creating three dimensional high goals point by point anatomical images
without the utilization of any harming radiations(Pradhan, 2010)[10]. The brain
MRI division into a few cerebrum tissues, for example, grey matter (GM), white
matter (WM) and cerebrospinal liquid (CSL) is exceedingly fundamental for the
conclusion of different ailments. This method is essentially used to distinguish the
itemized contrasts in the tissues in non-obtrusive style which have not been analyzed
by other imaging systems includes Computed Tomography (Sathies Kumar, 2017).
One of the serious issues in this entire procedure is isolating the anomalous cells
from the remainder of the image content which is known as the procedure of division.
The manual division is very testing just as tedious undertaking because of complex
structure of the cerebrum and nonappearance of well-characterized limits among
various brain tissues. In spite of the fact way toward it’s partitioning the ideal area
exceeding testing and confused however it has increased tremendous significance
and a few examinations have been led in improving the exactness of this assignment
(Freixenet et al., 2002)[8](Logeswari & Karnan, 2010).
The outcome from the diverse image division methods helpful in acquiring
highlights of divided tumor locale (Sathies Kumar, 2017). Various research work
has been done and few calculations are proposed with distinguishing position,
limit of tumors consequently so they can do assist conclusion at their soonest.
The examination exhibited in this work surveys the strategies and procedures of
programmed division brain tumor from the MRI images. The remainder work and
its areas are organized in the consequent way: In the second segment exhibits a
conventional strategy received in the procedures executed for brain tumor division,
trailed by a nitty gritty Literature overview in the third segment exhibits the general
examination and assessment of the outcomes pursued by the last ends and future
proposals in the closure segments.

63
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

3. EXISTING METHODOLOGY

In the existing work, medical image processing to extract features and segment
tumor from MRI images. It also dealt about Supervised, Unsupervised classification
techniques. Figure 2, Elaborates the steps involved in segmentation of MRI image.

3.1 Pre Processing

Producing images from various therapeutic imaging procedures acquire pointless

clamor into the image. A great deal of clamor accompanies in MRI, CT examine,
Mammographic Image or and so on. This commotion is an obstacle when dividing
tumor distinct from the given information image. To overcome this issue, first
preprocess the image so as to expel undesirable exceptions, and then send for further
preparing. In this preprocessing step deals procedures like clamor evacuation, channel
application, image upgrade, standardization, and so on.

Figure 2. Brain Tumor Detection Systems

3.2 Feature Extraction

Post handling step is needed when the image has been separated into segments, so
as to improve any edges and obscure any undesirable subtleties. This progression is
called include extraction where highlights from the image are removed for examination
that improved the tumor locale. In that most utilized component extraction steps
includes Morphological activities, edge discovery strategies or histogram leveling.

64
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

3.3 Segmentation Classifications

Separating a image into various portions to further translation and division of image
accomplished using numerous points of view. MR Images contain a high measure
of information that makes the assignment of deciphering difficult and repetitive for
a radiologist and clinical imaging authority. Similarly, the outcomes could diverse
relying on the experience of the specific authority (Khadem, 2010).Additionally,
various imaging frameworks present demand in the images, subsequently making it
hard to fragment cerebrum tumor and give an adequate presentation. The significant
division helps to compute the quantitative proportion of tumor in the cerebrum which
is basic for treatment of patient and follow up of the infection.
An objective point of huge number of PC vision, image preparing and AI based
applications recognized and separate the significant examples or indispensable
highlights from the image information described by the machine for additionally
gritty clarification (Tang et al., 2000)[7](Pradhan, 2010). The analysis of cerebrum
tumors and complex illnesses from radiographs is one of the most significant testing
because of high-time utilization and mutilation among these images help clinicians.
The fundamental point of different research gatherings exhibit solid calculations
that perform towards precise division. In this method, leads to develop a vigorous
just as to guarantee a protected conclusion framework.

3.4 Supervised Learning Method

In the programmed cerebrum tumor location, prepared the model utilizing the
regulated and solo AI strategies. As discussed above, many administered and
unaided procedures are accessible for underlying trials as we individually managed
solo AI methods utilizing CNN and K-Means. Additionally, the work utilized the
morphological administrator for our examinations and these systems are portrayed
underneath. The flow of brain tumor detection using supervised learning technique
discussed in sections 3.4.1

3.4.1 Convolutional Neural Network (CNN)

In computational sciences and in AI, the Convolutional neural systems (CNNs)

increased the acknowledgment for assortment of image handling applications
especially in programmed medicinal image division. As image segmentation isolates
a image into various parts, the target of segmentation is to plan straightforward
calculations as conceivable that process the images and examine the data in
progressively powerful, significant, reasonable and advantageous way. Utilizing CNN

65
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

designs as a model such significant data from medicinal images and radiographs
promptly removed out with increased exactness and improved execution time.
The discriminative model CNN legitimately gains from clarified images with no
earlier information (Liu, 2015). CNN based systems utilized the dataset preparation
to educate a system and these prepared systems anticipate the class names as well
as concentrate out the significant highlights includes designs, edges, lines. In that,
further train the other arrangement of classifiers and patches of the data extracted
out from the MRI images handled through convolution based channels. In that
method acquired the intricate highlights and help to yield the area, size of the
tumors dependent on their registered class scores. Besides, the CNN models have
additionally bit of programmed learning the perplexing highlights which identified
the solid tissues too strange tissues procured from MRI images.

3.4.2 Support Vector Machine (SVM)

SVM is one of the best predictable strategies for arranging the highlights. In SVM,
the arrangement of images is fundamentally partitioned into two different resultant
classes and the order is performed by finding the hyper-plane rule that separates
the two classes as shown in figure 3. SVM builds a hyper plane receiving a part
work (Sathies Kumar, 2017)[19] exhibited in the below figure3. In that, element
vectors on the left half of the primary hyper plane place with the class - 1 whereas
the component vectors assigned on the correct side of the fundamental hyper plane
compares to the class +1.

Figure 3. Hyper Plane Classifications

66
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

The segmentation with SVM predominantly relies on the following stages (a)
include extraction from preparing image (b) determination of SVM model (c)
readiness of informational index (d) SVM preparing a classification plane (Mengqiao
et al., 2017)[21]. A. Kumar et al. (Kumar, 2017) examination utilized Support vector
machine integrated with K-implies grouping and Principle Component Analysis for
extractions that order the tumor locale inside the cerebrum. In the given approach
information including cerebrum outputs was prepared utilizing bolster vector machine
whereas the tumor was sectioned utilizing k-means and PCA. The SVM classifier
discovers the class of the tumor identified and had an exactness of 96% for tumor
by the outcomes. Other than sectioning the tumor locale, the work additionally gave
a definite data on K-means and PCA between them. Additionally, G. Gupta et al.,
SVM combined with Fuzzy C-Means to order the images and their methodology to
apply FCM to segment the image and use of SVM to further group the images that
gave progressively upgraded and better outcomes (Gupta & Singh, 2017).

3.4.3 Bayesian Approach

In this methodology, information is accepted to pursue a multivariate ordinary

dispersion, where mean and covariance are evaluated from the preparation
informational index (Jobin Christ et al., 2009). The technique consolidates a chart
based calculation and Bayesian model portions the edema more over. Additionally it
tends to be stretched out to vectorial factors to work on multi-methodology images.
A Bayesian system is a model of compound likelihood circulation capacity of a lot of
variable like coordinated non-cyclic diagram with a likelihood table for every hub.
The hub in a Bayesian system relies on various factors in a space, and the circular
segments among hubs to the reliance connections among the factors (Jobin Christ
et al., 2009). The typical tissue classes are characterized by the enlisted spatial chart
book to the patient images and the tumor spatial earlier is determined from the
distinction image through histogram analysis (Corso et al., 2006). Also the likelihood
conveyance of tumor and edema has been thought to be an ordinary appropriation
that not right in the all cases (Corso et al., 2008). On account of edema, the creators
have expected a small amount of white issue likelihood for edema (Jagath, 2001)
incorporate the subsequent model-brain full affinities into the staggered division by
weighted accumulation calculation and apply the system undertake the partitioning
cerebrum tumor and edema in multichannel MR volumes.

67
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

3.5 Unsupervised Learning Technique

In the learning method, the groups are framed with class labels. Here, discussed
about K-implies grouping and morphological administrators that work as an unaided
learning strategy for brain tumor discovery.

3.5.1 Fuzzy C-Means Clustering

Grouping approach has been broadly utilized in various computational spaces

includes AI, PC vision and image handling. In addition, bunching method has late
advanced into different biomedical and social insurance applications prevalently for
the discovery of strange brain tissues (tumors) from radiographs procured through
attractive reverberation imaging (MRI) methodology (Benson, 2016). Dunn et al.,
presented a bunching based Fuzzy C-Means (FCM) approach, which extraordinarily
encouraged the division capability that bifurcates gathering of information into two
or different distinctive relating groups as displayed in Figure 4

Figure 4. Clustering with FVM.

In this strategy, the examination of cerebrum tumor division recommended

the method for FCM grouping calculation and the division of brain tumor bring
dynamic cells, necrotic center and edema (Martin et al., 2001). In Suganya et
al.,(2016) checked Fuzzy C Means calculation and its different applications in
restorative imaging, design acknowledgment, bioinformatics and information mining
(Suganya & Shanthi, 2012). Pham et. al., built up a novel grouping calculation by
incorporating fluffy entropy bunching for division of brain tissues (tumors) from

68
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

MRI images (Pham et al., 2018). Kumar et. al., proposed a changed intuitionistic
fluffy c-implies calculation (MIFCM) to systematically take care of the enhancement
issue utilizing lag range strategy for dubious multipliers. The proposed MIFCM
strategy concentrates cerebrum MRI information by defeating the confinements of
commotion and loose estimation (Kumar, 2018). Shanmuga Priya et. al., foreseen
FCM based staggered division by consolidating fluffy c-implies for distinguishing
the tumor tissues and edema among brain MRI images. The bunching procedure
improved by combining different bits dependent on the spatial data to perform
effective division (Priya, 2018).

3.5.2 K-Means Clustering

In a standard K- Means calculation, there are four stages includes: introduction,

grouping, computational and intermingling condition. The procedure introduces
by partitioning a specific informational into stable positive K number of bunches
with the goal that k centroids. The subsequent stage chooses a point which is in
colleague to a given informational collection and changes it to the nearest centroid.
The main gathering is finished when initial step is done which possibly happens
when no any point exist of last. The result of initial step brings about K new centroids
of the groups which ought to be recalculated toward the finish of the initial step.
When new K centroids are initiated, there is a need of another association among
similar informational collection focuses and the closest new centroid. In this way,
a circle is made which aides in examining the changed area of K centroids in each
stage except if all centroids come to static. As it were, centroids don’t move any
more appeared in Figure 5.

Figure 5. Clustering though K-means.

69
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

In Vijay et. al., (Vijay & Subhashini, 2013) contemplated the issue of marking in
image segmentation uncommonly with regards to robotized brain tumor identification.
They proposed a strategy that utilized morphological tasks for preprocessing and
K-implies procedure with a slight change that was to diminish the quantity of cycles
required for legitimate bunching by recommending the registered separation among
a group focus and information point under assessment, which is put away in an
information structure. This mix of K-implies bunching, and morphological tasks
created 95% precise outcomes on an example space of 100 MRI images, as delineated
in the outcomes. Dr. Patil et al. proposed a PC supported application to segmentation
tumor from the given MRI filters (Patil, 2005). The division thought embraced for
the investigation worked with an amalgamation of K-means bunching and Fuzzy
C means based grouping draws near. Four distinct modalities of images were tried
for investigations and the outcomes were produced dependent on parameters like,
Mean Square Error (MSE), Contrast, Correlation, Max blunder, Area, and so on.
The examination and results presumed that the technique proposed was strong,
precise and efficient.

3.5.3 Expectation Maximization

In the model based tumor segmentation method utilized an Expectation Maximization

(EM) separated the solid and the mild tissues. A lot of tumor attributes are exhibited
which is exceptionally fundamental for precise division. However, the work examined
the segment tumor area. EM steps are shown in the accompanying advances (Balafar
et al., 2010):

Step1: Initialize mean and Covariance framework utilizing K-Means.

Step2: Calculate participation likelihood of each preparation information.
Step3: Compute mean and fluctuation of each Gaussian part utilizing enrollment
capacity got in stage 2.

The stage 2 and 3 are rehashed until assembly. Gauss blend vector of each class
is derived by EM preparing information for that class. The utilizations of the EM
calculation brain MR image division accounted by (Wells et al., 1996) and (Leemput
et al., 1999).

4. COMPARATIVE STUDY AND DISCUSSION

In the examination on different segmentation calculations, MATLAB tool used

K-Means (Patil, 2005), Fuzzy C Means (Suganya & Shanthi, 2012) procedures whereas

70
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

Ibrahim et al. utilized CNN for order and exactness of the methods analyzed. As per
the outcomes, K-Means and Fuzzy C Means calculations had a similar precision.
G. Rao et al. (Kamnitsas et al., 2017) added to the space by differentiating Fuzzy
C-Means and K-Means bunching methods. In this segmentation through FCM and
K-Means was contrasted with Mean Square Error (MSE), Peak Signal to Noise Ratio
(PSNR), Peak Time (PTime) and region estimation. Consequently investigation
demonstrated that FCM had increased exactness of roughly 93%, alongside lower
PTime in contrast with K-Means is about 76% precision.
Image Segmentation is one of the most central ideas in PC vision. Segmentation
intends to change the representation of image to separate significant data from it.
This work displays a complete survey of various division techniques for cerebrum
tumor segmentation. Cross breed approaches integrated with various calculations
of these techniques had exhibited. The different robotized and semi-mechanized
procedures examined on constant usage contribute the greatness of PC innovation
to aid the field of therapeutic science. The examination recommends the regulated
learning technique have better exactness while inclination based strategies s are
precise and require lesser assets.

5. CONCLUSION

In this book chapter, various cerebrum tumor segmentation methods are surveyed
and also the steps for performing segmentation is explored and analytical results
for the same are analyzed with respect to the size and shape of images. This sort
of improvements in administered learning structures help in institutionalizing the
present strategies that help in clinical acknowledgment.

REFERENCES

Aggarwal, R., & Kaur, A. (2012). Comparative Analysis of Different Algorithms

For Brain Tumor Detection. International Journal of Scientific Research.
Balafar, M. A., Ramli, A. R., Saripan, M. I., & Mashohor, S. (2010). Review of
brain MRI segmentation methods. Artificial Intelligence Review, 33(3), 261–274.
doi:10.100710462-010-9155-0
Benson, C. C. (2016). Brain Tumor Segmentation from MR Brain Images using
Improved Fuzzy c-Means Clustering and Watershed Algorithm. 2016 Intl. Conference
on Advances in Computing, Communications and Informatics (ICACCI). 10.1109/
ICACCI.2016.7732045

71
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

Corso, J. J., Sharon, E., Brandt, A., & Yuille, A. (2006). Multilevel Segmentation
and Integrated Bayesian Model Classification with an Application to Brain Tumor
Segmentation. MICCAI, 4191, 790–798. doi:10.1007/11866763_97 PMID:17354845
Corso, J. J., Sharon, E., Dube, S., El-Saden, S., Sinha, U., & Yuille, A. (2008, May).
Efficient Multilevel Brain Tumor Segmentation with Integrated Bayesian Model
Classification. IEEE Transactions on Medical Imaging, 27(5), 629–640. doi:10.1109/
TMI.2007.912817 PMID:18450536
Dvorak, P., & Menze, B. (2015). Structured prediction with convolutional neural
networks for multimodal brain tumor segmentation. Proceeding of the Multimodal
Brain Tumor Image Segmentation Challenge, 13-24.
Freixenet, J., Munoz, X., Raba, D., Marti, J., & Cufi, X. (2002). Yet another survey
on image segmentation: Region and boundary information integration. Proc. 7th
Eur. Conf. Computer Vision Part III, 408–422. 10.1007/3-540-47977-5_27
Girshick, R. (2014). Rich feature hierarchies for accurate object detection and
semantic segmentation. Proceedings of the IEEE conference on computer vision
and pattern recognition. 10.1109/CVPR.2014.81
Gupta & Singh. (2017). Brain Tumor segmentation and classification using Fcm
and support vector machine. International Research Journal of Engineering and
Technology, 4(5).
Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C.,
Jodoin, P.-M., & Larochelle, H. (2017). Brain tumor segmentation with deep neural
networks. Medical Image Analysis, 35, 18–31. doi:10.1016/j.media.2016.05.004
PMID:27310171
Jagath, C. (2001, October). Bayesian Approach to Segmentation of Statistical
Parametric Maps. IEEE Transactions on Biomedical Engineering, 48(10).
Jobin Christ, M. C., Sasikumar, K., & Parwathy, R. M. S. (2009, July). Application
of Bayesian Method in Medical Image Segmentation. International Journal of
Computing Science and Communication Technologies, VOL, 2(1).
Kamnitsas, K., Ledig, C., Newcombe, V. F. J., Simpson, J. P., Kane, A. D., Menon,
D. K., Rueckert, D., & Glocker, B. (2017). Efficient multi-scale 3D CNN with fully
connected CRF for accurate brain lesion segmentation. Medical Image Analysis,
36, 61–78. doi:10.1016/j.media.2016.10.004 PMID:27865153
Kaur & Rani. (2016). MRI Brain Tumor Segmentation Methods- A Review.
International Journal of Current Engineering and Technology.

72
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

Khadem. (2010). MRI Brain image segmentation using graph cuts (Master’s thesis).
Chalmers University of Technology, Goteborg, Sweden.
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with
deep convolutional neural networks. Advances in Neural Information Processing
Systems.
Kumar, A. (2017). A Novel Approach for Brain Tumor Detection Using Support
Vector Machine. K-Means and PCA Algorithm.
Kumar, D. (2018). A modified intuitionistic fuzzy c-means clustering approach to
segment human brain MRI image. Multimedia Tools and Applications, 1–25.
Laddha, R. R. (2014). A Review on Brain Tumor Detection Using Segmentation And
Threshold Operations. International Journal of Computer Science and Information
Technologies, 5(1), 607–611.
Leemput, K. V., Maes, F., Vandermeulen, D., & Suetens, P. (1999). Automated
model-based tissue classification of MR images of brain. IEEE Transactions on
Medical Imaging, 18(10), 897–908. doi:10.1109/42.811270 PMID:10628949
Liu, Z. (2015). Semantic image segmentation via deep parsing network. Proceedings
of the IEEE International Conference on Computer Vision. 10.1109/ICCV.2015.162
Logeswari & Karnan. (2010). An improved implementation of brain tumor detection
using segmentation based on soft computing. Journal of Cancer Research and
Experimental Oncology, 2(1).
Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for
semantic segmentation. Proceedings of the IEEE conference on computer vision
and pattern recognition.
Martin, D., Fowlkes, C., Tal, D., & Malik, J. (2001). A database of human segmented
natural images and its application to evaluating segmentationalgorithms and
measuring ecological statistics. Proc. 8th Int. Conf. Computer Vision, 2, 416–423.
10.1109/ICCV.2001.937655
Mengqiao, W., Jie, Y., Yilei, C., & Hao, W. (2017). The multimodal brain tumor
image segmentation based on convolutional neural networks. 2017 2nd IEEE
International Conference on Computational Intelligence and Applications (ICCIA),
336-339. 10.1109/CIAPP.2017.8167234
Moon, N., Bullitt, E., Leemput, K., & Gerig, G. (2002). Model based brain
and tumor segmentation. Int. Conf. on Pattern Recognition, 528-531. 10.1109/
ICPR.2002.1044787

73
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

Patil. (2005). Pachpande, Automatic Brain Tumor Detection Using K-Means.

Academic Press.
Pereira, S., Pinto, A., Alves, V., & Silva, C. A. (2016). Brain tumor segmentation
using convolutional neural networks in MRI images. IEEE Transactions on Medical
Imaging, 35(5), 1240–1251. doi:10.1109/TMI.2016.2538465 PMID:26960222
Pham, T. X., Siarry, P., & Oulhadj, H. (2018). Integrating fuzzy entropy clustering
with an improved PSO for MRI brain image segmentation. Applied Soft Computing,
65, 230–242. doi:10.1016/j.asoc.2018.01.003
Pradhan, S. (2010). Development of Unsupervised Image Segmentation Schemes for
Brain MRI using HMRF model (Master Thesis). Department of EE, NIT, Rourkela,
India.
Priya. (2018). Efficient fuzzy c-means based multilevel image segmentation for brain
tumor detection in MR images. Design Automation for Embedded Systems, 1–13.
Sathies Kumar, T. (2017). Brain Tumor Detection Using SVM Classifier. IEEE
3rd International Conference on Sensing, Signal Processing and Security (ICSSS).
Suganya, R., & Shanthi, R. (2012). Fuzzy C- Means Algorithm. RE:view, 2(11), 1–3.
Swapnil, R. T. (2016). Detection of brain tumor from MRI images by using
segmentation & SVM. World Conference on Futuristic Trends in Research and
Innovation for Social Welfare (Startup Conclave).
Tang, H., Wu, E. X., Ma, Q. Y., Gallagher, D., Perera, G. M., & Zhuang, T. (2000).
MRI brain image segmentation by multi-resolution edge detection and region selection.
Computerized Medical Imaging and Graphics, 24(6), 349–357. doi:10.1016/S0895-
6111(00)00037-9 PMID:11008183
Thuy, Hai, & Thai. (n.d.). Image Classification using Support Vector Machine and
Artificial Neural Network. Academic Press.
Urban, G. (2014). Multi-modal brain tumor segmentation using deep convolutional
neural networks. MICCAI BraTS (Brain Tumor Segmentation) Challenge.
Proceedings, 31-35.
Vijay, J., & Subhashini, J. (2013). An Efficient Brain Tumor Detection Methodology
Using K-Means Clustering Algorithm. International conference on Communication
and Signal Processing.
Wang. (2017). The multimodal brain tumor image segmentation based on
convolutional neural networks. ICCIA.

74
Detection of Tumor From Brain MRI Images Using Supervised and Unsupervised Methods

Wells, W. M., Grimson, W. E. L., Kikinis, R., & Jolesz, F. A. (1996). Adaptive
segmentation of MRI data. IEEE Transactions on Medical Imaging, 15(4), 429–442.
doi:10.1109/42.511747 PMID:18215925
Yi, D. (2016). 3-D convolutional neural networks for glioblastoma segmentation.
arXiv preprint arXiv:1611.04534.
Zheng, S. (2015). Conditional random fields as recurrent neural networks. Proceedings
of the IEEE international conference on computer vision. 10.1109/ICCV.2015.179
Zikic, D. (2014). Segmentation of brain tumor tissues with convolutional neural
networks. Proceedings MICCAI-BRATS, 36-39.
Zulpe & Chowhan. (2011). Statical Approach For MRI Brain Tumor Quantification.
International Journal of Computer Applications, 35(7).

75
76

Chapter 4
Breast Cancer Diagnosis
in Mammograms Using
Wavelet Analysis, Haralick
Descriptors, and Autoencoder
Maira Araujo de Santana
Universidade Federal de Pernambuco, Brazil

Jessiane Mônica Silva Pereira

Universidade de Pernambuco, Brazil

Washington Wagner Azevedo da Silva

Universidade Federal de Pernambuco, Brazil

Wellington Pinheiro dos Santos

https://orcid.org/0000-0003-2558-6602
Universidade Federal de Pernambuco, Brazil

ABSTRACT
In this chapter, the authors used autoencoder in data preprocessing step in an attempt
to improve image representation, consequently increasing classification performance.
The authors applied autoencoder to the task of breast lesion classification in
mammographic images. Image Retrieval in Medical Applications (IRMA) database
was used. This database has a total of 2,796 ROI (regions of interest) images from
mammograms. The images are from patients in one of the three conditions: with a
benign lesion, a malignant lesion, or presenting healthy breast. In this study, images
were from mostly fatty breasts and authors assessed different intelligent algorithms
performance in grouping the images in their respective diagnosis.

DOI: 10.4018/978-1-7998-3092-4.ch004

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

INTRODUCTION

Cancer is a leading cause of death and, nowadays, is one of the largest public health
issue worldwide. For decades, breast cancer has been the most common type of
cancer among women around the world. The World Health Organization (WHO)
estimates an occurrence of 1.7 million new cases per year (DeSantis et al., 2014).
This disease is now placed on the top five causes of cancer death around the world
(American Cancer Society, 2019). Survival rates for breast cancer can range from
80%, in high-income countries, to below 40%, in low-income countries (Coleman
et al., 2008). The low survival rate in some countries is due to the lack of early
detection programs. These programs have a major impact on the success of cancer
treatment, since treatment becomes more difficult in later stages.
The gold standard method for breast cancer diagnosis is the digital mammography
(Maitra, Nag & Bandyopadhyay, 2011). However, visual analysis of mammography
can be a difficult task, even for specialists. Imaging diagnosis is a complex task
due to the great variability of clinical cases (Ferreira, Oliveira & Martinez, 2011).
Most of the cases observed in clinical practice do not match to classical images and
theoretical descriptions (Juhl, Crummy, & Kuhlman, 2000). That is why Computer
Aided Diagnosis (CAD) plays an important role in helping radiologists to improve
diagnosis accuracy.
Many studies worldwide, are applying traditional image processing and
analysis techniques to medical field. Therefore, the combination of professionals
specialized knowledge and pattern recognition computational tools may improve
diagnosis accuracy (Araujo et al., 2012; Azevedo et al., 2015; Bandyopadhyay,
2010; Commowick et al., 2018; Cordeiro, Bezerra & Santos, 2017; Cordeiro et al.,
2012; Cordeiro, Santos & Silva-Filho, 2013; Cordeiro, Santos & Silva-Filho, 2016a;
Cordeiro, Santos & Silva-Filho, 2016b; Cruz, Cruz e Santos, 2018; Fernandes &
Santos, 2014; Lima, Silva-Filho & Santos, 2014; Mascaro et al., 2009; Santana et
al., 2017; Santos, Assis, Souza & Santos Filho, 2009; Santos et al., 2008a; Santos
et al., 2008b; Santos et al., 2009a; Santos et al., 2009b; Santos et al., 2010; Santos,
Souza & Santos Filho, 2017). Intelligent systems may be used to assist these
professionals in decision-making, thus improving the efficiency in identifying
anatomical abnormalities (Araujo et al., 2012; Azevedo et al., 2015; Commowick
et al., 2018; Cordeiro, Bezerra & Santos, 2017; Cordeiro et al., 2012; Cordeiro,
Santos & Silva-Filho, 2013; Cordeiro, Santos & Silva-Filho, 2016a; Cordeiro,
Santos & Silva-Filho, 2016b; Cruz, Cruz e Santos, 2018; Fernandes & Santos, 2014;
Ferreira, Oliveira & Martinez, 2011; Lima, Silva-Filho & Santos, 2014; Mascaro et
al., 2009; Santana et al., 2017; Santos, Assis, Souza & Santos Filho, 2009; Santos
et al., 2008a; Santos et al., 2008b; Santos et al., 2009a; Santos et al., 2009b; Santos

77
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

et al., 2010; Santos, Souza & Santos Filho, 2017; Wang, Yuan & Sheng, 2010; Ye,
Zheng & Hao, 2010).
This chapter proposes the use of autoencoders to optimize images representation.
As a case of study, authors applied the method to the task of detecting and classifying
lesions in regions of interest of mammograms. They compared the results to previous
approaches, also using Haralick descriptors, Wavelet transform and intelligent
classifiers.

BACKGROUND

In this session, authors provide some related works and a broad definition of some
topics they used along the experiments.

Related Works

In the study of Abdel-Zaher and Eldeib (2016) they developed a CAD approach for
breast cancer detection. They used deep belief network unsupervised path followed by
back-propagation supervised path. They proposed a neural back-propagation network
with the Liebenberg Marquardt learning function. The weights are initialized from
the deep belief network path (DBN-NN). They used the Wisconsin Breast Cancer
Dataset (WBCD) to assess technique performance. The complex classifier achieved
an accuracy of 99.68%, indicating promising results, when compared to previously
published studies. The proposed system provides an effective classification model
for breast cancer. In addition, we examined the architecture in several pieces of
training-testing.
Bayramoglu, Kannala and Heikkila (2016) aimed to identify breast cancer using
histopathological images, independent of their extensions using convolutional neural
networks (CNNs). They proposed two different architectures: a single task CNN
was used to predict malignancy and a multitasking CNN was used to simultaneously
predict malignancy and the level of image enlargement. They used BreaKHis
database to evaluate and compare the results to previous ones. The results of the
experiments showed that the proposed approach improved the performance of the
specific magnification model, regardless of magnification. Even though having
an limited set of training data, the obtained results with the proposed model are
comparable to previous results obtained by the state-of-the-art and results obtained
by handmade resources. However, unlike previous methods, the proposed approach
has the potential to directly benefit from additional training data. Such additional
data can be captured at magnification levels equal to or different from previous data.

78
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

In the paper of Khuriwal and Mishra, they propose applying the convolutional
neural network deep learning algorithm to the diagnosis of breast cancer. They used
the mammography MIAS database. Their study shows how we can use deep learning
technology to diagnose breast cancer using the MIAS Dataset.
In this work, when applying the Deep Learning technology in the database, an
accuracy of 98% was achieved. MIAS database provides 200 images and 12 features
in the data set. In this study, they used 12 features, which were extracted after
preprocessing. However, before the training model, some preprocessing algorithms,
such as Watershed Segmentation, Color based segmentation and Adaptive Mean
Filters to staggered datasets were applied. After that, they applied the proposed
model to perform classification. In this study, the Deep Learning algorithm is also
compared to other machine learning algorithms. They found that the presented
methodology achieves better results than other widely used intelligent algorithms.
In Jannesari et al. (2018), they applied pre-trained and adjusted Deep Learning
networks. First, the authors tried to discriminate between different types of cancer.
They used 6,402 tissue microarray samples (TMAs). Models, including ResNet V1
50, correctly predicted 99.8% of the four types of cancer, including breast, bladder,
lung and lymphoma. In a second moment, they tried to assess method performance
for the classification of breast cancer subtypes. To do so, they used 7,909 images
of 82 patients, from the BreakHis database. ResNet V1 152 classified benign and
malignant breast cancers with an accuracy of 98.7%. In addition, ResNet V150 and
ResNet V1 152 categorized in benign (adenoses, fibroadenoma, lodia and tubular
adenoma) or malignant (ductal carcinomas, lobular carcinomas, mucinous carcinomas
and papillary carcinomas) subtypes with 94.8% and 96.4% accuracy, respectively.
Confusion matrices revealed high sensitivity values of 1, 0.995 and 0.993 for cancers,
as well as malignant and benign subgroups, respectively. The scores of the areas
under the curve (AUC) were 0.996 for cancers, 0.973 for malign subtype and 0.996
for benign subtype. One of the most significant and impressive results to emerge
from the data analysis was the insignificant false positive (FP) and false negative
(FN). The optimal results indicate that FP is between 0 and 4 while FN is between
0 and 8 on which test data including 800, 900, 809, 1000 for four given classes.
In the studies pointed out by Xiao et al. (2018), they show a new method,
integrating an unsupervised features extraction algorithm based on deep learning.
They combined stacked autoencoders to a support vector machine, thus creating the
SAE-SVM model. The approach was applied for breast cancer diagnosis. Stacked
autoencoders with a fast pre-training layer and an improved momentum refresh
training algorithm are applied to acquire essential information and extract relevant
features from the original data. Next, they used a support vector machine to classify
samples with new features into malignant and benign lesions. They tested the proposed
method using the Wisconsin Diagnostic Breast Cancer database. Performance was

79
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

assessed using various measures and compared to previously published results. The
results shows that the proposed SAE-SVM method improves accuracy to 98.25%
and outperforms other methods. The unsupervised features extraction based on
deep learning significantly improves the classification performance and provides a
promising approach to breast cancer diagnosis.

Features Extraction

As mentioned before, we used Haralick descriptors and Wavelet transform for features
extraction. The first one extracts texture-based features from statistical calculations
between neighboring pixels of the image (Haralick, Shanmugam & Dinstein, 1973).
These features are obtained through a co-occurrence matrix. Co-occurrence matrix
has the color occurrence values at a given image and represents the spatial distribution
and dependence of gray levels within a local area. Haralick features are widely used
as image descriptors, and had also been applied for breast lesion detection (Azevedo
et al., 2015; Bhateja et al., 2018; Jenifer, Parasuraman & Kadirvel, 2014; Kishore et
al., 2014; Santana et al., 2018; Yasiran, Salleh & Mahmud, 2016). From Haralick
descriptor it is possible to differentiate textures that do not follow a certain pattern
of repetition in the image (Haralick, Shanmugam & Dinstein, 1973).
Wavelets, however, are very effective tools for representing multi-resolution
images. The wavelet transform relative to image processing can be implemented in
a two-dimensional way. Mallat proposed a Discrete Wavelet Transform of a signal
through the decomposition of an original image into a series of images generated by
discrete high-pass and low-pass filters (Mallat, 1999). Such as Haralick descriptors,
Wavelet transform is being successfully exploited to mammography representation,
in order to detect breast lesions (Eltoukhy, Faye & Samir, 2009; Ganesan et al., 2014;
Joseph & Balakrishnan, 2011; Roberts et al., 2017).

Autoencoder

Autoencoder is a neural network that is trained so that the input number is equal to
the output number. Its main purpose is optimize the representation of the input data.
It has an unsupervised training, so there is no need of labeled data. This training is
based on optimizing a cost function. The cost function consists on the mean square
error metric, which measures the error between the input, x, and its reconstruction
in the output, y (Maria et al., 2016; Xu & Zhang, 2015).
The autoencoder architecture consists of an encoder and a decoder. The encoder
structures the input data. Soon after, the decoder reverts the structuring to reconstruct
the original input. The output layer has the same number of neurons as the input layer.
It is done in order to reconstruct its own inputs, without predicting its outputs, by

80
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Table 1. Autoencoder configuration

Parameter Value
Hidden Size 10
Encoder Transfer Function logsig
Decoder Transfer Function logsig
Max Epochs 1000
L2 Weight Regularization 0.001
Loss Function msesparse
Sparsity Proportion 0.05
Sparsity Regularization 1

using the unsupervised training (Vincent et al., 2008). Table 1 shows the parameters
we set for the autoencoder to perform the experiments presented in this chapter.
The Hidden Size parameter matches the number of neurons in the hidden layer.
Encoder Transfer Function and Decoder Transfer Function represent the transfer
function for the encoder and decoder, respectively. In this application, the function
used for both encoder and decoder, is the logsig, which is described by Equation 1.

1
f (z ) = (1)
1 + e −z

Max Epochs parameter denes the number of interaction. The coefficient L2

Weight Regularization is the regularizer. The Loss Function denes cost function and
in this case it stands for to the mean square error. Sparsity Proportion controls the
dispersion of the hidden layer output. Finally, the Sparsity Regularization parameter
controls the dispersion smoothing impact on the cost function (MathWorks, 2019).

PROPOSED METHOD

In this study, we used Image Retrieval in Medical Applications (IRMA) database

(Deserno et al., 2012a; Deserno et al., 2012b; Oliveira, 2010). This database has
2,796 ROI (Regions of Interest) images from mammograms, which were classified
by radiologists and were resized to 128x128 pixels. IRMA was developed in
Aachen University of Technology (RWTH Aachen), in Germany, and results from
a combination of four open access repositories:

81
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

• 150 images from the Mini-MIAS database (Suckling et al., 1994);

• 2,576 images from DDSM (Digital Database for Screening Mammography)
(Heath, Bowyer & Kopans, 2000);
• 1 image from the LLN database (Lawrence Livermore National Laboratory);
• 69 images from the Department of Radiology at Aachen University of
Technology (RWTH), Germany.

The images from IRMA database have four types of tissue density, classified
according to the BI-RADS classification (D’Orsi et al., 2013) into: adipose tissue
(Type I), fibrous tissue (Type II), heterogeneously dense tissue (Type III), and
extremely dense tissue (Type IV). Figure 1 shows examples of each of these classes.
Moreover, IRMA has images of breasts with malignant lesion, benign lesion and of
healthy breasts, in which there is no lesion.

Figure 1. Mammograms of different breast tissues: (a) adipose tissue, (b) fibrous
tissue, (c) heterogeneously dense tissue and (d) extremely dense tissue.
Source: The authors

For this study, authors considered only the images of fatty breasts (Type I). They
chose to use this class because most women who undergo mammography have this
tissue composition in their breasts, since the amount of adipose tissue in the breasts
tends to increase with age. Samples of these images can be seen in Figure 2.
Authors access to IRMA database was allowed upon an agreement between
Federal University of Pernambuco, Brazil (UFPE) and the Department of Medical
Informatics of Aachen University of Technology, Germany. This agreement vetoes
the commercial use of partial or whole database.
In this chapter, the authors propose to use autoencoders to preprocess features
from mammographic images, in order to optimize database representation. The
features used in this study were extracted using both Haralick texture extractor and
Wavelet transform. Our set of image features was submitted to the encoder and
decoder processes, which worked as a kind of filter.

82
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Figure 2. Example of ROI images from IRMA database classified as Type I.

Source: The authors

After features extraction and preprocessing, authors finally assessed classification

performance of some widely used algorithms. The classification step was conducted
using: Bayes Net, Naive Bayes, Multi-Layer Perceptron (MLP), Support Vector
Machine (SVM), J48, Random Tree, Random Forest, Extreme Learning Machine
(ELM) and Morphological Extreme Learning Machine (mELM).
K-fold cross-validation method was used to perform all experiments. The authors
also repeated each experiment 30 times, in order to acquire statistical information
from them.

SOLUTIONS AND RECOMMENDATIONS

In this section, authors present the results obtained for the classification of breast
images in one of the three possible diagnosis: benign lesion, malignant lesion or
healthy breast.
Since the main goal in this study was to compare the quality of database
representation, authors conducted all experiments in two different databases: the first
one without using autoencoders and the second one using autoencoder to preprocess
the features. Tables 2 and 3 show the mean and standard deviation (STD) for both
accuracy and kappa statistic achieved for each classifier.
In Table 2 are the results for the database created without using autoencoder.
From this table, you may see that mELM with erosion kernel outperformed the
other methods in terms of both accuracy and kappa. mELM with erosion achieved
an average accuracy of 93.28% and 0.92 of kappa. It was closely followed by SVM
with linear kernel, mELM with dilatation kernel and ELM. MLP, Random Forest,

83
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

J48 and SVM with polynomial kernel achieved less satisfying performance, all close
to 80% of accuracy. Bayesian classifiers performed worse than the other methods,
with accuracy around 60% and kappa around 0.45. Regarding to the standard
deviation, mELMs and ELM achieved the best results, which were equal or really
close to 0 (zero), meaning a very low data dispersion. Greater standard deviation
values were associated to Naive Bayes and Random Tree classifiers, reaching a
maximum value of 5.82.

Table 2. Classification results without using autoencoder

Classifier Parameter Accuracy (%) Kappa Statistic

Bayes Net - 64.08 ± 5.12 0.46 ± 0.08
Naive Bayes - 61.66 ± 5.82 0.42 ± 0.09
MLP - 84.20 ± 5.04 0.76 ± 0.08
Linear kernel 92.29 ± 3.05 0.88 ± 0.05
SVM
Poly kernel (P=2) 78.27 ± 5.02 0.67 ± 0.08
J48 - 78.45 ± 4.89 0.68 ± 0.07
Random Tree - 71.11 ± 5.61 0.57 ± 0.08
Random Forest 100 trees 81.46 ± 4.47 0.72 ± 0.07
ELM Sigmoid kernel 91.86 ± 0.77 0.91 ± 0.01
Dilatation kernel 92.13 ± 0.00 0.91 ± 0.00
mELM
Erosion kernel 93.28 ± 0.00 0.92 ± 0.00

Table 3 presents the results for the database in which we used autoencoder.
These results shows that the use of autoencoder in the preprocessing step triggered
a decrease on classifiers performance. While authors achieved an accuracy of
93.28% in the previous dataset, they found a maximum of 78.08% of accuracy for
the dataset with autoencoder. The maximum accuracy, in this case (78.08%), was
achieved by SVM classifier with polynomial kernel. However, this algorithm did not
reached the best result for kappa statistic in this scenario. Again, the best kappa, of
0.73, was achieved by the mELMs with both kernels. Regarding to accuracy, SVM
was followed by mELMs, Random Forest, MLP and ELM. One more time, Bayes
Net and Naive Bayes were associated to the worst performances overall. Bayesian
classifiers also presented the low values for kappa statistic, remaining close to 0.40.
As to data dispersion, we observe a small decrease in the maximum value for standard
deviation after using autoencoder. However, the methods that showed to minimize
dispersion in the previous dataset (mELMs and ELM), presented an increase of up
to 5.29 in standard deviation when using autoencoder.

84
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Table 3. Classification results using autoencoder

Classifier Parameter Accuracy (%) Kappa Statistic

Bayes Net - 61.29 ± 5.03 0.42 ± 0.07
Naive Bayes - 59.89 ± 5.70 0.40 ± 0.08
MLP - 75.36 ± 5.07 0.63 ± 0.07
Linear kernel 76.44 ± 5.40 0.65 ± 0.08
SVM
Poly kernel (P=2) 78.08 ± 4.87 0.67 ± 0.07
J48 - 70.77 ± 5.34 0.56 ± 0.08
Random Tree - 67.31 ± 5.41 0.51 ± 0.08
Random Forest 100 trees 75.86 ± 4.61 0.64 ± 0.06
ELM Sigmoid kernel 75.01 ± 4.84 0.72 ± 0.05
Dilatation kernel 76.24 ± 4.88 0.73 ± 0.05
mELM
Erosion kernel 76.13 ± 5.29 0.73 ± 0.06

CONCLUSION

This chapter proposed the application of autoencoder in mammographic images

preprocessing step. The authors aimed to perform a kind of image filtering, in an
attempt to build a better representation of the images and improve classification
performance. Tools such as the autoencoder are being exploited to be used in
optimization of pattern recognition systems. However, their findings did not meet
expectations. Instead of improving classification, the use of autoencoders worsened
the performance of the algorithms. They believe that the preprocessing method
reduced the image details responsible for characterizing each breast lesions, worsening
image representation.
It is important mentioning that the results shown in this chapter does not invalidates
the use of autoencoders for identification of breast lesion in mammograms. The
authors have no intention of generalizing the results presented here by stating that
autoencoder should not be applied in situations like this. In fact, they actually aim
to further analyze these results and possibly take a closer look on the parameters of
this method. For future studies, the authors believe that they may improve results
by modulating autoencoder parameters and characteristics. When it comes to
diagnostic solutions, it is worth to keep on investing in optimization methods such
as autoencoders, so that, one day, we may achieve the best possible performance.

85
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

ACKNOWLEDGMENT

We thank “Fundação de Amparo à Pesquisa do Estado de Pernambuco”, FACEPE,

Brazil, and “Conselho Nacional de Desenvolvimento Científico e Tecnológico”,
CNPq, Brazil, for the partial financial support for this research.

REFERENCES

Abdel-Zaher, A. M., & Eldeib, A. M. (2016). Breast cancer classification using deep
belief networks. Expert Systems with Applications, 46(1), 139–144. doi:10.1016/j.
eswa.2015.10.015
American Cancer Society. (2019). Cancer Facts & Figures 2019. American Cancer
Society.
Araujo, M., Queiroz, K., Pininga, M., Lima, R., & Santos, W. (2012). Uso de regiões
elipsoidais como ferramenta de segmentação em termogramas de mama. In XXIII
Congresso Brasileiro de Engenharia Biomédica (CBEB 2012). Pernambuco: SBEB.
Azevedo, W. W., Lima, S. M., Fernandes, I. M., Rocha, A. D., Cordeiro, F. R.,
Silva-Filho, A. G., & Santos, W. P. (2015). Fuzzy Morphological Extreme Learning
Machines to Detect and Classify Masses in Mammograms. In 2015 IEEE International
Conference on Fuzzy Systems. IEEE. 10.1109/FUZZ-IEEE.2015.7337975
Bandyopadhyay, S. K. (2010). Survey on Segmentation Methods for Locating Masses
in a Mammogram Image. International Journal of Computers and Applications,
9(11), 25–28. doi:10.5120/1429-1926
Bayramoglu, N., Kannala, J., & Heikkila, J. (2016). Deep Learning for Magnification
Independent Breast Cancer Histopathology Image Classification. In 2016 23rd
International Conference on Pattern Recognition (ICPR). Cancun: IEEE. 10.1109/
ICPR.2016.7900002
Bhateja, V., Gautam, A., Tiwari, A., Bao, L. N., Satapathy, S. C., Nhu, N. G., & Le,
D.-N. (2018). Haralick Features-Based Classification of Mammograms Using SVM.
In Information Systems Design and Intelligent Applications (Vol. 672). Springer.
doi:10.1007/978-981-10-7512-4_77

86
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Coleman, M. P., Quaresma, M., Berrino, F., Lutz, J. M., De Angelis, R., Capocaccia,
R., Baili, P., Rachet, B., Gatta, G., Hakulinen, T., Micheli, A., Sant, M., Weir, H.
K., Elwood, J. M., Tsukuma, H., Koifman, S., & Silva, E. (2008). Cancer survival
in five continents: A worldwide population-based study (CONCORD). The Lancet.
Oncology, 9(8), 730–756. doi:10.1016/S1470-2045(08)70179-7 PMID:18639491
Commowick, O., Istace, A., Kain, M., Laurent, B., Leray, F., Simon, M., Pop, S.
C., Girard, P., Ameli, R., Ferré, J.-C., Kerbrat, A., Tourdias, T., Cervenansky, F.,
Glatard, T., Beaumont, J., Doyle, S., Forbes, F., Knight, J., Khademi, A., ... Barillot,
C. (2018). Objective Evaluation of Multiple Sclerosis Lesion Segmentation Using
a Data Management and Processing Infrastructure. Scientific Reports, 8(1), 13650.
doi:10.103841598-018-31911-7 PMID:30209345
Cordeiro, F. R., Bezerra, K. F. P., & Santos, W. P. (2017). Random walker with
fuzzy initialization applied to segment masses in mammography images. In 30th
International Symposium on Computer-Based Medical Systems (CBMS). IEEE.
10.1109/CBMS.2017.40
Cordeiro, F. R., Lima, S. M., Silva-Filho, A. G., & Santos, W. P. (2012). Segmentation
of mammography by applying extreme learning machine in tumor detection. In
International Conference of Intelligent Data Engineering and Automated Learning.
Berlin: Springer.
Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2013). Segmentation of
mammography by applying growcut for mass detection. Studies in Health Technology
and Informatics, 192, 87–91. PMID:23920521
Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2016a). A semi-supervised fuzzy
growcut algorithm to segment and classify regions of interest of mammographic images.
Expert Systems with Applications, 65, 116–126. doi:10.1016/j.eswa.2016.08.016
Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2016b). An adaptive semi-
supervised fuzzy growcut algorithm to segment masses of regions of interest of
mammographic images. Applied Soft Computing, 46, 613–628. doi:10.1016/j.
asoc.2015.11.040
Cruz, T. N., Cruz, T. M., & Santos, W. P. (2018). Detection and classification of
lesions in mammographies using neural networks and morphological wavelets.
IEEE Latin America Transactions, 16(3), 926–932. doi:10.1109/TLA.2018.8358675
D’Orsi, C. J., Sickles, E. A., Mendelson, E. B., & Morris, E. A. (2013). Breast
Imaging Reporting and Data System: ACR BI-RADS breast imaging atlas (5th ed.).
American College of Radiology.

87
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

DeSantis, C. E., Lin, C. C., Mariotto, A. B., Siegel, R. L., Stein, K. D., Kramer, J.
L., Alteri, R., Robbins, A. S., & Jemal, A. (2014). Cancer treatment and survivorship
statistics, 2014. CA: a Cancer Journal for Clinicians, 64(4), 252–271. doi:10.3322/
caac.21235 PMID:24890451
Deserno, T., Soiron, M., Oliveira, J., & Araújo, A. (2012a). Towards Computer-Aided
Diagnostics of Screening Mammography Using Content-Based Image Retrieval. In
2011 24th SIBGRAPI Conference on Graphics, Patterns and Images. Alagoas: IEEE.
Deserno, T. M., Soiron, M., Oliveira, J. E. E., & Araújo, A. A. (2012b). Computer-
aided diagnostics of screening mammography using content-based image retrieval.
In Medical Imaging: Computer-Aided Diagnosis 2012. SPIE. doi:10.1117/12.912392
Eltoukhy, M., Faye, I., & Samir, B. (2009). Breast Cancer Diagnosis in Mammograms
using Multilevel Wavelet Analysis. Proceeding of National Postgraduate Conference.
Fernandes, I., & Santos, W. (2014). Classificação de mamografias utilizando
extração de atributos de textura e redes neurais artificiais. In Congresso Brasileiro
de Engenharia Biomédica (CBEB 2014). SBEB.
Ferreira, J., Oliveira, H., & Martinez, M. (2011). Aplicação de uma metodologia
computacional inteligente no diagnóstico de lesões cancerígenas. Revista Brasileira
de Inovação Tecnológica em Saúde, 1(2), 4-9.
Ganesan, K., Acharya, U. R., Chua, C. K., Min, L. C., & Abraham, T. K. (2014).
Automated Diagnosis of Mammogram Images of Breast Cancer Using Discrete
Wavelet Transform and Spherical Wavelet Transform Features: A Comparative
Study. Technology in Cancer Research & Treatment, 13(6), 605–615. doi:10.7785/
tcrtexpress.2013.600262 PMID:24000991
Haralick, R. M., Shanmugam, K., & Dinstein, I. (1973). Textural Features for Image
Classification. IEEE Transactions on Systems, Man, and Cybernetics, 3(6), 610–621.
doi:10.1109/TSMC.1973.4309314
Heath, M., Bowyer, K. W., & Kopans, D. (2000). The digital database for screening
mammography. Proceedings of the 5th International Workshop on Digital
Mammography
Jannesari, M., Habibzadeh, M., Aboulkheyr, H., Khosravi, P., Elemento, O.,
Totonchi, M., & Hajirasouliha, I. (2018). Breast Cancer Histopathological Image
Classification: A Deep Learning Approach. In 2018 IEEE International Conference
on Bioinformatics and Biomedicine (BIBM). IEEE. 10.1109/BIBM.2018.8621307

88
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Jenifer, S., Parasuraman, S., & Kadirvel, A. (2014). An Efficient Biomedical Imaging
Technique for Automatic Detection of Abnormalities in Digital Mammograms.
Journal of Medical Imaging and Health Informatics, 4(2), 291–296. doi:10.1166/
jmihi.2014.1246
Joseph, S., & Balakrishnan, K. (2011). Local Binary Patterns, Haar Wavelet Features
and Haralick Texture Features for Mammogram Image Classification using Artificial
Neural Networks. In International Conference on Advances in Computing and
Information Technology. Springer. 10.1007/978-3-642-22555-0_12
Juhl, J. H., Crummy, A. B., & Kuhlman, J. E. (2000). Paul & Juhl Interpretação
Radiológica (7a ed.). Rio de Janeiro: Guanabara-Koogan.
Khuriwal, N., & Mishra, N. (2018). Breast Cancer Detection from Histopathological
Images using Deep Learning. In 2018 3rd International Conference and Workshops
on Recent Advances and Innovations in Engineering (ICRAIE). IEEE. 10.1109/
ICRAIE.2018.8710426
Kishore, B., Arjunan, R. V., Saha, R., & Selvan, S. (2014). Using Haralick Features
for the Distance Measure Classification of Digital Mammograms. International
Journal of Computers and Applications, 6(1), 17–21.
Lima, S. M., Silva-Filho, A. G., & Santos, W. P. (2014). A methodology for
classification of lesions in mammographies using Zernike moments, ELM and SVM
neural networks in a multi-kernel approach. In 2014 IEEE International Conference
on Systems, Man, and Cybernetics (SMC). IEEE. 10.1109/SMC.2014.6974041
Maitra, I. K., Nag, S., & Bandyopadhyay, S. K. (2011). Identification of Abnormal
Masses in Digital Mammography Images. International Journal of Computer
Graphics, 2(1).
Mallat, S. (1999). A Wavelet Tour of Signal Processing. Academic.
Maria, J., Amaro, J., Falcao, G., & Alexandre, L. A. (2016). Stacked Autoencoders
Using Low-Power Accelerated Architectures for Object Recognition in Autonomous
Systems. Neural Processing Letters, 43(2), 445–458. doi:10.100711063-015-9430-9
Mascaro, A. A., Mello, C. A., Santos, W. P., & Cavalcanti, G. D. (2009).
Mammographic images segmentation using texture descriptors. In 2009 Annual
International Conference of the IEEE Engineering in Medicine and Biology Society.
IEEE. 10.1109/IEMBS.2009.5333696
MathWorks. (2019). Deep Learning Toolbox ™ Reference. Author.

89
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Oliveira, J. E., Machado, A. M., Chavez, G. C., Lopes, A. P., Deserno, T. M., &
Araújo, A. A. (2010). MammoSys: A content-based image retrieval system using
breast density patterns. Computer Methods and Programs in Biomedicine, 99(3),
289–297. doi:10.1016/j.cmpb.2010.01.005 PMID:20207441
Roberts, T., Newell, M., Auffermann, W., & Vidakovic, B. (2017). Wavelet-based
scaling indices for breast cancer diagnostics. Statistics in Medicine, 36(12), 1989–
2000. doi:10.1002im.7264 PMID:28226399
Santana, M., Pereira, J., Lima, N., Sousa, F., Lima, R., & Santos, W. (2017).
Classificação de lesões em imagens frontais de termografia de mama a partir de
sistema inteligente de suporte ao diagnóstico. In Anais do I Simpósio de Inovação
em Engenharia Biomédica. SABIO.
Santana, M. A., Pereira, J. M. S., Silva, F. L., Lima, N. M., Sousa, F. N., Arruda,
G. M. S., Lima, R. C. F., Silva, W. W. A., & Santos, W. P. (2018). Breast cancer
diagnosis based on mammary thermography and extreme learning machines.
Research on Biomedical Engineering, 34(1), 45–53. doi:10.1590/2446-4740.05217
Santos, W. P., Assis, F., Souza, R., Santos Filho, P. B., & Neto, F. L. (2009b).
Dialectical Multispectral Classification of Diffusion-weighted Magnetic Resonance
Images as an Alternative to Apparent Diffusion Coefficients Maps to Perform
Anatomical Analysis. Computerized Medical Imaging and Graphics, 33(6), 442–460.
doi:10.1016/j.compmedimag.2009.04.004 PMID:19446434
Santos, W. P., Assis, F. M., Souza, R. E., Mendes, P. B., Monteiro, H. S. S., & Alves,
H. D. (2009a). A Dialectical Method to Classify Alzheimer’s Magnetic Resonance
Images. In Evolutionary Computation. IntechOpen. doi:10.5772/9609
Santos, W. P., Assis, F. M., Souza, R. E., Mendes, P. B., Monteiro, H. S. S., &
Alves, H. D. (2010). Fuzzy-based Dialectical Non-supervised Image Classification
and Clustering. International Journal of Hybrid Intelligent Systems, 7(2), 115–124.
doi:10.3233/HIS-2010-0108
Santos, W. P., Assis, F. M., Souza, R. E., & Santos Filho, P. B. (2008a). Evaluation of
Alzheimer’s Disease by Analysis of MR Images using Objective Dialectical Classifiers
as an Alternative to ADC Maps. In 2008 30th Annual International Conference of
the IEEE Engineering in Medicine and Biology Society. IEEE.
Santos, W. P., Assis, F. M., Souza, R. E., & Santos Filho, P. B. (2009). Dialectical
Classification of MR Images for the Evaluation of Alzheimer’s Disease. In Recent
Advances in Biomedical Engineering. IntechOpen. doi:10.5772/7475

90
Breast Cancer Diagnosis in Mammograms Using Wavelet Analysis

Santos, W. P., Souza, R. E., & Santos Filho, P. B. (2017). Evaluation of Alzheimer’s
Disease by Analysis of MR Images using Multilayer Perceptrons and Kohonen SOM
Classifiers as an Alternative to the ADC Maps. In 2017 29th Annual International
Conference of the IEEE Engineering in Medicine and Biology Society. IEEE.
Santos, W. P., Souza, R. E., Santos Filho, P. B., Neto, F. B. L., & Assis, F. M.
(2008b). A Dialectical Approach for Classification of DW-MR Alzheimer’s Images.
In 2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on
COmputational Intelligence). Hong Kong: IEEE. 10.1109/CEC.2008.4631023
Suckling, J., Parker, J., Dance, D., Astley, S., Hutt, I., Boggis, C., Ricketts, I.,
Stamatakis, E., Cerneaz, N., Kok, S., Taylor, P., Betal, D., & Savage, J. (1994).
The mammographic image analysis society digital mammogram database. In 2nd
International Workshop on Digital Mammography. Excerpta Medica.
Vincent, P., Larochelle, H., Bengio, Y., & Manzagol, P.-A. (2008). Extracting and
Composing Robust Features with Denoising Autoencoders. In 25th International
Conference on Machine Learning. New York: ACM. 10.1145/1390156.1390294
Wang, D., Yuan, F., & Sheng, H. (2010). An Algorithm for Medical Imaging
Identification based on Edge Detection and Seed Filling. In 2010 International
Conference on Computer Application and System Modeling (ICCASM 2010).
Taiyuan: IEEE.
Xiao, Y., Wu, J., Lin, Z., & Zhao, X. (2018). Breast Cancer Diagnosis Using an
Unsupervised Feature Extraction Algorithm Based on Deep Learning. In 2018 37th
Chinese Control Conference (CCC). IEEE. 10.23919/ChiCC.2018.8483140
Xu, Q., & Zhang, L. (2015). The Effect of Different Hidden Unit Number of Sparse
Autoencoder. In The 27th Chinese Control and Decision Conference (2015 CCDC).
IEEE. 10.1109/CCDC.2015.7162335
Yasiran, S. S., Salleh, S., & Mahmud, R. (2016). Haralick texture and invariant
moments features for breast cancer classification. In AIP Conference Proceedings.
AIP Publishing. doi:10.1063/1.4954535
Ye, S., Zheng, S., & Hao, W. (2010). Medical image edge detection method based
on adaptive facet model. In 2010 International Conference on Computer Application
and System Modeling (ICCASM 2010). Taiyuan: IEEE.

91
92

Chapter 5
Feature Selection Using
Random Forest Algorithm
to Diagnose Tuberculosis
From Lung CT Images
Beaulah Jeyavathana Rajendran
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

Kanimozhi K. V.
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

ABSTRACT
Tuberculosis is one of the hazardous infectious diseases that can be categorized by
the evolution of tubercles in the tissues. This disease mainly affects the lungs and also
the other parts of the body. The disease can be easily diagnosed by the radiologists.
The main objective of this chapter is to get best solution selected by means of modified
particle swarm optimization is regarded as optimal feature descriptor. Five stages
are being used to detect tuberculosis disease. They are pre-processing an image,
segmenting the lungs and extracting the feature, feature selection and classification.
These stages that are used in medical image processing to identify the tuberculosis.
In the feature extraction, the GLCM approach is used to extract the features and
from the extracted feature sets the optimal features are selected by random forest.
Finally, support vector machine classifier method is used for image classification.
The experimentation is done, and intermediate results are obtained. The proposed
system accuracy results are better than the existing method in classification.

DOI: 10.4018/978-1-7998-3092-4.ch005

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

INTRODUCTION

Tuberculosis is one of the communicable bacterial diseases that is caused by the

bacterium called Mycobacterium tuberculosis and that may affect any tissues of the
body but it primarily disturbs the lungs. TB is one of the airborne pathogens that
can binge through air or by coughing or sneezing from one person to another. TB
disturbs all age groups in all parts of the world. Tuberculosis bacteria are present
in sputum trials and it is identified under a microscope.
In 2015, around 11 million people fell ill with TB and 2 million people were died
from the disease. Over 95% of the deaths in TB occur in low and middle- revenue
countries.
X-ray is not easily predicting the early stage of tuberculosis. Hence, because
of this wrong prediction of tuberculosis, an automated detection of tuberculosis is
used. To overcome the problems in existing methods, CT lung images are used for
diagnosis of tuberculosis.
In image processing Feature extraction is an important step, which is a special
form of dimensionality reduction. When the input data is too large to be processed
and alleged to be redundant then the data is transformed into a reduced set of
feature representations. Feature contains the information that is related to colour,
shape, texture and context. Modified Random Forest Algorithm technique is based
on optimization searching technique and it is used to find the optimal solutions. It
is used for selecting the best features after the feature extraction process. This will
continue until a needed solution is obtained. Classifying the images whether it is
normal or abnormal by SVM classifier.

RELATED WORKS

Les Folio (2014), presented the automated approach for detecting tuberculosis in
conventional poster anterior chest radiographs. For the extracted region, set of texture
features and shape features are computed, which enable the X-rays to be classified
as normal or abnormal using a binary classifier. The preprocessing techniques is
used to remove the noises and the feature extraction are done to extract the useful
features in given image and the feature selection technique will optimize the top
ranking features that are relevant for the image and the classifiers are employed to
classify the images and the performance measures are found for the same (Sun et al.,
2015). Laurens Hogeweg, Clara I. The performance is evaluated on a TB screening
and a TB suspect database using both an external and a radiological reference
standard. The systems to detect different types of TB related abnormalities and
their combination is described. Yan Kang (2015), Wenbo Li, using a new adaptive

93
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

VOI selection method. The improved GA algorithm to select the optimal feature
combination from the feature pool to establish SVM classifier (Omisore, 2014).
G. Vijaya, A. Suhasini identified the cancer tumor from lung CT images using
edge detection and boundary tracing. To classify the lung cancer, by using the data
mining, classification techniques like SMO (Sequential Minimal Optimization),
J48 decision tree, Naive Bayes.
Once the classification is performed, we have to compare the experimental results
of the above classification techniques, and determine which one gives accurate and
correct answers (Girisha et al., 2013). Mumini Olatunji Omisore (2014) proposed
the genetic neuro-fuzzy inferential model for the diagnosis of tuberculosis. Finally,
SVM is used in the classification stage (Linguraru et al., n.d.). A. Zabidi, L.Y.
Khuan IEEE International conference(2011) proposed the Binary Particle Swarm
Optimization For Feature selection in Detection of Infants with Hypothyroidism. In
this, he investigates the effect of feature selection with Binary PSO on performance
of Multilayer perceptron classifier in discriminating between the healthy infants and
infants with hypothyroidism from their cry signals. The performance was examined
by varying the number of coefficients.

PROPOSED WORK

Material and Methods

In the study, dataset containing lung CT images comprising abnormal lung and
normal lung are taken from several patients was utilized. The lung diseases are
categorized by the radiologist from the CT Image. Images are collected from male
and female patients whose ages are ranging from 15 to 78 years.

Pre-Processing

Pre-processing is done to remove unwanted noise and it gives quality to the images
at this stage where filtering is done to remove noise. In our proposed system we
have used wiener filter to remove noise. Wiener filter preserves the edges and fine
details of lungs. It is low pass-filter. The filter size of 5*5 is selected to avoid over
smoothing of the image. 2D Wiener filter is used for lessening of additive gaussian
white noise in images.

94
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

SEGMENTATION

K-Means Clustering

K-Means is possibly the most well-known clustering algorithm. It’s easy to understand
and implement in code.

1. To begin, we first hand-picked a number of classes/groups to practice and

erratically set their respective center points. To number out the number of
classes to use, it’s good to take a rapid look at the data and attempt to recognize
any distinct groupings. The center points are vectors of the same length as
each data point vector.
2. Each data point is classified by calculating the distance between that point
and each group center, and then classifying the point to be in the group whose
center is closest to it.
3. Based on these classified points, we recompute the group center by taking the
mean of all the vectors in the group.
4. Repeat these steps for a set number of iterations or until the group centers don’t
change much between iterations. We can also opt to randomly initialize the
group centers a few times, and then select the run that looks like it provided
the best results.

ROI EXTRACTION

ROI’s are taken out using the radiologists and thus it is authorized to attain the clinical
relevance which progresses the performance of the system. Extract the defected
tissues from the lung as ROI’s and then find the intensity level of the pixels and
using the range of pixel intensity values discriminate the defected tissues and other
lung tissues. If there is no defected tissues are present, then the slice is considered
to be Normal. Then obtain the class labels for each ROI’s from the experts. Finally,
ROI’s are extracted and also the class label information is obtained.

FEATURE EXTRACTION

GLCM Approach

The feature extraction based on Texture feature is carried out. GLCM approach is
used for extracting the features in given image such as entropy, energy, contrast,

95
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

correlation, variance, sum average, homogeneity cluster shade and etc.., are
considered for feature selection. Extract these twenty-two features for each ROI in
four orientations 0o, 45o, 90o, 135o using GLCM also called the Grey Tone Spatial
Dependency Matrix. GLCM contains the information about the positions of pixel
having similar grey level values.

CLASSIFICATION SUBSYSTEM

Naïve Bayes Classifier

It is not a single algorithm but a family of algorithms where all of them stake a
communal principle, i.e. every pair of features being classified is self-determining
of each other. Naïve Bayes classifiers are highly ascendable, necessitating a number
of parameters linear in the number of variables (features/predictors) in a learning
problem.

FEATURE SELECTION

The term Feature selection deals with selecting a subset of features, among the entire
features, that shows the best performance in classification accuracy. Optimization
searching process is done by Modified Random Forest Algorithm.

Modified Random Forest Algorithm

Random forests are an unification of tree predictors such that each tree depends on
the values of a random vector sampled self-reliantly. In year 2001, Leo Breiman, a
statistician recognizes the difficulties in prevailing machine learning techniques. In
former tree approach of machine learning data set is not consistently disseminated
lead to imbalance of data. Imbalanced data set performance is underprivileged
with the classification, this lead to miss classification and error in the training
phase. He recommended data set were collected and then divided into two or more
subset of data, where one or more data set used as learner and residual is used for
test purpose. Many researchers got fascinated towards Random Forest approach of
handling data set and started employed on different attributes of Random Forest
like features, concepts, analysis and modification of the proposed model of Random
Forest algorithm. Research works going on in the field of Random Forest can be
approximately classified into three categories:

96
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

• Research in Random Forest Improving accuracy

• Improving performance
• Investigation of new domain for application of Random Forest.

Number of trees engendered in the Random Forest is the task for the Researchers
because it devours extra space in memory and also increases run time of the algorithm.
Rudimentary problem with this method is that it takes only time complexity under
contemplation leaving the space complexity. Because of this pruning approach is
required. Redeemable more time when assessment to static approach but tactlessly
it is hard to contrivance because of this researchers are also not showing interest
for this approach. Research work completed under static pruning approach fall in
to three majority categories:

• Weighted voting method

• Ranking Based method
• Search Based method

In one forerunner work of static pruning, genetic algorithm is used to choice

most optimal candidate from pool of Decision Tree. Other work uses elimination of
similar Decision Tree if their output class and accuracy are same then keep single
copy of Tree eliminating others. Dynamic pruning necessitates help of statistics and
probability along with nature inspired algorithm to get recovering results. In one
approach authors have tried to model dynamic pruning approach with the help of
eight degree mathematical equation.

RESULTS AND DISCUSSION

The CT images used for testing and training purpose for classification were collected
from AARTHI SCANS & LABS at TIRUNELVELI. We have several CT images,
but we use 197 images for my work out of which 94 images have tuberculosis and
the remaining images do not have tuberculosis. The segmentation of the image
takes place, in which K-Means Clustering is done. The set of Tuberculosis (TB) CT
images and non-Tuberculosis CT images are tested to give an accurate result. Thus,
the technique deals with the accurate detection of tuberculosis.

97
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

Figure 1.

Processing Time Analysis

In our study, to implement our proposed algorithm, we used MATLAB software

(R2016a) on a laptop, Intel Core i3 (2.0 GHZ) and 4GB memory. The resolution of
images in our database was 512 x 512.
To evaluate our proposed algorithm efficiently, we analyzed each step of our
algorithm based on processing time. Table 1 presents average of the processing time
of each module of proposed algorithm.

98
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

Table 1.

Proposed algorithm. Processing time of each Module

Module Processing time (s)
Pre-processing 1.1456 seconds
K-Means segmentation 20.0196 seconds
Feature extraction 0.24534 seconds

Table 2.

Using Modified Random

Parameter Using PSO Using MPSO
Forest
TP 43 46 49
TN 41 45 48
FP 09 05 02
FN 07 04 01
Accuracy 84 91 98
Sensitivity 86 92 96
Specificity 82 90 94

Table 3.

Performance
Classifier Percentage
Parameters
Accuracy 92.30%
Bayes
Sensitivity 96%
classifier
Specificity 49%

CONCLUSION

In this work the preprocessing of the images is done, then the segmentation are done
by K-Means Clustering algorithm. It is one of the distinctive clustering algorithms.
Furthermore several algorithms are developed based on K-Means. During the
implementation of this algorithm, find some points that can be further improvement
in the future using some advanced clustering to achieve more accuracy and GLCM-
based feature extraction technique was described. The texture features are served as
the input to classify the image accurately. Effective use of these multiple features and
the selection of suitable classification method is significant for improving accuracy.

99
Feature Selection Using Random Forest Algorithm to Diagnose Tuberculosis

REFERENCES

Bhuvaneswari, Aruna, & Loganathan. (2014). Classification of Lung Diseases by

Image Processing Techniques Using Computed Tomography Images. International
Journal of Advanced Computer Research, 4(1).
Candemir, Jaeger, Palaniappan, & Musco. (2014). Lung Segmentation in Chest
Radiographs Using Anatomical Atlases With Non-rigid Registration. IEEE
Transactions on Medical Imaging, 33(2).
Dai, S., Lu, K., & Dong, J. (2015). Lung segmentation with improved graph cuts
on chest CT images. 3rd IAPR Asian Conference on Pattern Recognition. 10.1109/
ACPR.2015.7486502
Girisha, Chandrashekhar, & Kurian. (2013). Texture Feature Extraction of Video
Frames Using GLCM. International Journal of Engineering Trends and Technology,
4(6).
Hogeweg, Sánchez, Maduskar, Philipsen, & Story. (2015). Automatic Detection
of Tuberculosis in Chest Radiographs Using a Combination of Textural, Focal and
Shape Abnormality Analysis. IEEE Transactions on Medical Imaging, 34(12).
Jaeger, Karargyris, Candemir, Folio, & Siegelman. (2014). Automatic Tuberculosis
Screening Using Chest Radiographs. IEEE Transactions on Medical Imaging, 33(2).
Linguraru, Richbourg, Liu, & Watt. (n.d.). Tumor Burden Analysis on Computed
Tomography by Automated Liver and Tumor Segmentation. IEEE.
Omisore. (2014). Proposed the genetic neuro-fuzzy inferential model for the diagnosis
of tuberculosis. IEEE Transactions.
Rendon-Gonzalez & Ponomaryov. (2016). Automatic Lung Nodule Segmentation
and Classification in CT Images Based on SVM. International conferences IEEE.
Sun, S., Li, W., & Kang, Y. (2015). Lung Nodule Detection Based on GA and
SVM. 8th International Conference on Bio Medical Engineering and Informatics
(BMEI 2015).
Vijaya, Suhasini, & Priya. (n.d.). Automatic detection of lung cancer in CT images.
IJRET: International Journal of Research in Engineering and Technology.

100
101

Chapter 6
An Ensemble Feature Subset
Selection for Women Breast
Cancer Classification
A. Kalaivani
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

ABSTRACT
Breast cancer leads to fatal diseases both in India and America and takes the lives
of thousands of women in the world every year. The patients can be easily treated if
the signs and symptoms are identified at the early stages. But the symptoms identified
at the final stage spreads in the human body, and most of the time, the cancer is
identified at the final stage. Breast cancer detected at the early stage is treated easily
rather than at the advanced stage. Computer-aided diagnosis came into existence
from 2000 with high expectations to improve true positive diagnosis and reduce false
positive marks. Artificial intelligence revolved in computing drives the attention of
deep learning for an automated breast cancer detection and diagnosis in digital
mammography. The chapter focuses on automatic feature selection algorithm for
diagnosis of women breast cancer from digital mammographic images achieved
through multi-layer perceptron techniques.

1. INTRODUCTION

Breast cancer (BC) is the tumor that originates in the cells of women breast and grows
into breast cancer. Breast Cancer tumor has a nature to spread to different parts of
the body (Y.S. Hotko, 2013). Breast Cancer is a universal disease which harms the

DOI: 10.4018/978-1-7998-3092-4.ch006

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

lives of women in the age group of 25– 50. There is a potential rise in the number
of BC cases in India and America. During the past five years, the survival rates
of BC patients are about 90% in the USA and whereas in India the figure reports
approximately 60%. Breast Cancer projection for India suspect to reach higher rates
may be two millions (S. Malvia, 2017).
Medical World identified hormonal, life style and environmental factors are
the root cause for development of Breast Cancer. Around 5%–6% of breast cancer
patients are due to gene mutations that went through the ages of the family. The
most common factors due to which breast cancer caused are Obesity, increasing
age, postmenopausal hormonal imbalances. The only mechanism to diagnose breast
cancer The early detection of breast cancer can reduce the costs of the treatment as
there is no prevention mechanism for breast cancer. But the early detection is difficult
since most of the times it is unusual to show cancer symptoms. It is indispensable
for the patients to test using digital mammograms or self-breast tests to detect any
early irregularities in the breast and also to get the tumor advanced (Shallu, Rajesh
Mehra, 2018).
Medical Experts only deals with the diagnosis of disease purely based on the
various tests performed upon the patient. The important factors in diagnosis is
based on the data evaluation of patients data and experts knowledge. The medical
diagnosis focused on this paper leads to the early diagnosis of women breast cancer
from digital mammographic images predicts the malignant cases in a timely manner
and which increased life span of patients from 56 to 86%.
Breast Cancer shows four signs of liaisons which are micro-calcification, mass,
architectural distortion, and breast asymmetries(Hazlina H,et.al., 2004). The medical
modalities supported for breast cancer diagnosis are positron emission tomography
(PET), magnetic resonance imaging (MRI), CT scan, X-ray, digital mammography,
ultrasound, tomography of photo-acoustic, optical, electrical impedance, opto-
acoustic imaging(Sulochana Wadhwani et.al., 2013). The results obtained from these
methods are used to recognize the patterns, which help medical experts to classify
breast cancer into malignant or benign cases.
Digital Mammography System used for early stage breast cancer replaces X-ray
film by electronics produces mammographic pictures of the breast enables better
picture with a lower radiation dose. The breast images are transferred to a computer
for review by the radiologist and can also be used for long term storage of patient
record.
As per World Health Organization report, women breast cancer found to be
the common women diagnosed cancer disease which also leads to death mortality
among women worldwide. On an average, a woman is diagnosed with breast cancer
every two minutes and one woman dies of it every 13 minutes worldwide. Survey
statistics in 2019, says an estimated 2,68,600 new cases of invasive breast cancer

102
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Figure 1. Digital mammography system

(from (Bronzino, 2000)).

are diagnosed in U.S women and 62,960 new cases of noninvasive breast cancer
and mortality rate reached 41,760.
iThe recent introduction of slide scanners that digitize the biopsy into multi-
resolution images, along with advances in deep learning methods are used for
computer-aided diagnosis of breast cancer. The intermediate steps from tissue
localization, image enhancement, segmentation, annotation made us to make
diagnosis accurate, reliable, efficient and cost-effective.
The introduction of deep learning convolutional neural networks (CNNs) in
medical image analysis has brought forth a potential revolution in computer-based
interpretation of Digital Mammography. Deep learning convolutional neural networks
involve the processing of an image by multiple sequential stages of maximum
convolutions and down sampling operators which combine the spatially correlated
information contained in images. During this multiple-stage process, this information
is broken down into different representations, and the analysis are more abstract,
and the ability of the network to recognize the image was made accurate.
The paper introduces the discussed the related work by the researchers for women
breast cancer using computer aided detection and diagnosis system in in section 2.
A detailed view of how an Artificial Neural Network system can play a vital role
in CAD diagnosis and the proposed system methodology is explained in section 3.
Materials and Methods of the proposed technology are explained in Section 5 and
experimental results and discussion are done in Section 6 and finally paper is ended
in the conclusion section.

2. RELATED WORK

The Computer Aided Diagnosis system detects the suspicious regions with high
sensitivity and presents the results to the radiologist with a focus to reduce false
positives. The preprocessing algorithm reduces the noise acquired in the image.

103
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

A segmentation process will identifies different region of interest to detect high

suspicion of some signs of cancer. Features are extracted from the segmented
region and the classification of positives or negatives prediction of Breast Cancer
is obtained through an artificial neural network.
A brief breakup of different types of artificial neural network architecture, which
is done by a number of researchers are shown in table 2. Computer Aided diagnosis
reduce the workloads on clinicians by detecting artifacts and providing decision support
for better performance. Table I provides information about different artificial neural
network algorithms studied by various researchers to diagnose women breast cancer.
Researchers have carried out the research work on the artificial neural network
techniques and applied various algorithms to diagnose women breast cancer. List
of the researchers works are listed and explained in the table 2.

3. ARTIFICIAL NEURAL NETWORKS

An artificial neural network (ANN) is a computational model based on the basis of

biological neural networks similar representation of human neural system. ANN
consists of an interconnected group of artificial neurons and processes information
using a connectionist approach to computation. Artificial Neural Network architectures
are designed to model complex relationships between inputs and outputs or used
to find patterns in data.

Table 1. Research work on Medical Images based on Artifineural Neural Network

Image
S.Nos. ANN Types Research Outcome
Modality
1 Cellular Neural Network Detect Boundary/ area X-ray
2 GA and CNN Detect nodular shadows X-ray
3 Hybrid Neural Digital CAD Classify 3-15 mm size nodules X-ray
4 ANN Feed Forward Increase sensitivity & accuracy X-ray
5 Artificial CNN & application Detect False Positive & increase sensitivity X-ray
6 Convolution Neural Network Decrease False & Increase True Positive X-ray
Two - level Convolution Neural
7 Reduce False Positive X-ray
Network
8 NN Ensembles Reduce False Positive X-ray
9 J-net Improve sensitivity & accuracy CT Image
10 Massive Training ANN Enhancement of lung nodules CT Image

104
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Table 2. Existing methods of neural network algorithms for Women Breast Cancer
Diagnosis

S.Nos. Authors Technique Algorithm Results

Classification of Breast
cancer into malignant or
Sulochana Wadhwani,
Artificial Neural Back propagation benign with the accuracies
1 A.K Wadhwani, Monika
Network Algorithm of 94.11% and 100%
Saraswat.
(Sulochana Wadhwani et.
al., 2013)
Computer Aided
Pankaj Sapra, Detection of Brain Tumor,
Detection System and Competitive
2 Rupinderpal Singh, obtained 100% accuracy
Probabilistic Neural Learning Algorithm
Shivani Khurana. (Pankaj Sapra et.al, 2013)
Network
Yongjun WU, Na Wang,
Back propagation Provides accuracy of
Hongsheng ZHANG, Artificial Neural
3 Algorithm Diagnosis 96.6%. (Yongjun WU et.
Lijuan Qin, Zhen YAN, Network
of lung cancer. al., 2010)
Yiming WU.
Classification of
Ayoub Arafi, Youssef Image processing Multilayer mammographic images of
4 Safi, Rkia Fajr and and Artificial Neural Perceptron Training breast cancer. Accuracy
Abdelaziz Bouroumi. Network Algorithm obtained is 95.49%.
(Ayoub Arafi et. al., 2013)
Adaptive Resonance
Seema Singh, Sunita Artificial Neural Obtained accuracy 82.64%
5 Theory Detection of
Saini, Mandeep Singh. Network (Seema Singh et. al., 2012)
cancer using ART
Back propagation
Algorithm Breast provided an accuracy of
Ali Raad, Ali Kalakech, . Artificial Neural
6 cancer detection and 94% (Ali Raad et. al.,
Mohammad Ayache. Network
classification using 2012)
ANN.
Breast cancer detection
using hierarchical RBF
Yuehui Chen, Yan Wang, Artificial neural Hierarchical Radial
7 with the accuracy of
Bo Yang. network Basis Function
97.09%.
(Yuehui Chen et. al., 2006)

Artificial Neural Network are composed of multiple nodes present in input layer,
hidden layer and output layer similar to nerve system of human body. The nodes in
the input layer, hidden layer and output layer interact with each other which can take
input data and perform operations on the input nodes. The result of these operations
is passed to other nodes in the next layer and the output at each node are activated to
the next layer through an activation function. Weights are associated in the node for
learning the network and final output is obtained in the output layer. The artificial
neural network architecture is of two broad categories: feed forward neural network
and feed backward neural network.

105
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Figure 2. Artificial Neural Network

A feed forward neural network is an artificial neural network connections between

the nodes are forwarded from input layer to hidden layer and forwarded from hidden
layer into output layer. The feed forward neural network was the first and simplest
type of artificial neural network devised which will not form loops. The information
in feed forward neural network moves in single forward direction, moving from the
input nodes, to the hidden nodes and finally move to the output nodes. There are no
cycles or loops in the network. f’(x)=f(x)(1-f(x))}.The Feed forward network can
be formed using a single layer perceptron and multi-layer perceptron.

Figure 3. Feed Forward Neural Network

The most popular form of artificial neural network architecture focused by the
researchers for effective medical diagnosis is the multilayer perceptron (MLP). A
multilayer perceptron artificial neural network has any number of inputs, which can
have one or more hidden layers with any number of units and takes forward sigmoid

106
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

activation functions in the hidden layers. MLP is highly suitable to approximate a

classification function which is determined by the vector attribute values based on
one or more classes. MLP trained with forward or backward propagation algorithm
can be effectively used for medical diagnosis.
In our proposed research work we have used WEKA tool for the feature selection
and training the model using multi-layer perceptron. WEKA is an open source software
stands for Waikato Environment for Knowledge Learning which was developed
by the University of Waikato, New Zealand. WEKA tool supports for data mining
and machine learning algorithms tasks such as data preprocessing, classification,
clustering, regression, feature selection and visualize the data. The data formats to be
handled by WEKA are ARFF, CSV, C4.5 and binary and can also port the data from
URL or an SQL database. After loading the data, preprocessing filters could be used
for adding/removing features, attributes, discretization, sampling, randomizing etc.
Classifier model can be implemented using training and testing model building
using multilayer perceptron neural network. For model building, the entire data
women breast cancer data set are split into 70% training and 30% testing groups.
The preprocessed feature subset datasets are applied into this training model and
the data can be again built using Cross Validation 0f 10-fold. The training data sets
includes 9 folds and the test set includes a single fold. The Multilayer Perceptron
neural network is build and the classifier performance are analyzed. The datasets for
different feature subsets generated are observed and the final results are analyzed
for to fix the model for further medical analysis. Unnecessary features are removed
from the data set using filter ranking method and MLP neural network training
model is built which gives a better result than simple train & test method. For the
most fare evaluation of the classifier model result of K-fold cross validation method
of MLP training was used.
Multi-Layer Perceptron neural network classifier creates two sets training and
test set which are used for MLP neural network training in WEKA. Training sets
and Test sets are used for learning which fits the weight parameters of the classifier.
Test set are the examples used only to assess the performance of a fully-specified
trained classifier. The algorithm for MLP neural network classifier model to train
and test is given in table 3.
The learning rate and momentum are adjusted to get a better training result. Main
parameters for building the classifier with a number of hidden layers, learning rate,
momentum, trainingTime. Paramenter setting function isweka.classifiers.functions.
MultilayerPerceptron
For better classifier model building and testing, the 10-fold Cross Validation is
used in which datasets are randomly divided into 10 folds in which any one fold is
used as a testing set and the remaining 9 subsets are used as training sets.

107
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Table 3. MLP Neural Network Classifier

Multi-Layer Perceptron Neural Network Classifier Model

Initialize all weights to small random numbers Until satisfied DO
For each training example, do
1. Input the training example to the network and compute the training outputs
2. For each output unit k

δk ← ok (1 − ok )(tk − ok )

3. For each hidden unit h

δh ← oh (1 − oh ) ∑ wh ,k δk
k ∈outputs

4. Update each network weight

wi, j ← wi, j + ∆wi, j where

Learning rate Momentum

↑ ↑
∆wi, j = η δj x i . j ∆wi, j (n ) = ηδj x i . j + α ∆wi, j (n − 1)

4. MATERIALS AND METHODS

Computer Aided Diagnosis systems perform automatic assessments of patient images

and present to radiologist towards the appearance of an abnormality. Computer Aided
Diagnosis is an effective and efficient solution for implementing double reading,
which provides double perception and interpretation. In a Real Time Scenario,
technicians can be attracted by some features and can miss a lesion which can be used
to identify disease. The CAD systems read images faster without reducing accuracy
than radiologists, but medical community needs to be confidant in the results.
The automatic computer aided diagnosis of breast cancer is an important real-world
medical problem and an active research field in the detection of subtle abnormalities in
mammograms to improve the accuracy and efficiency of radiologists. Mammograms
are the images but difficult to interpret, and a mammographic images are preprocessed
to improve the quality of the images and make the feature extraction phase more
reliable. The image enhancement technique is applied to enhance the quality of the
images. After preprocessing phase, features relevant to the classification are extracted
from the preprocessed images in which the most relevant and non-redundant features
are selected for the classification model to develop.

108
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

The proposed work data sets are taken from UCI machine repository. It consists
of 569 Fine Needle Aspirate biopsy samples of human breast tissues. There are 32
attributes computed for each cell samples which includes radius, perimeter, texture,
area, smoothness, compactness, concavity, concave points, symmetry and fractal
dimension are the 10 most important features which have been used as the only
inputs to the network as these are sufficient to obtain good results. This makes the
network more concise and less complex.
We used the WEKA toolkit to experiment the breast cancer dataset captured
through digital mammography used to evaluate the performance and effectiveness
of the breast cancer prediction models. The features of the given women breast
cancer data sets features radius, texture, perimeter, area, smoothness, compactness,
concavity, concave points, symmetry, fractal dimensions of mean data. The features
includes radius, texture, perimeter, area, smoothness, compactness, concavity, concave
points, symmetry, fractal dimensions of standard error data. The worst data features
of radius, texture, perimeter, area, smoothness, compactness, concavity, concave
points, symmetry, fractal dimensions and finally includes class label for classification.
The various Feature Selection Methods chosen such as fssubset evaluation, filtered
subset evaluation, Gain Ratio, Chisquare, SVM Attribute, Relief Attribute with their
corresponding search methods such as Best first and Ranker methods. The features
selected are using Gain Ratio, InfoGain, Chisquare, Filtered Attribute Evaluation,
One R Attribute Evaluation, Relief Attribute Evaluator, Symmetrical Uncertain
Evaluator using Ranking Methods are given below in the table 4.
Features subsets are selected based on the filtered methods and list of the features
are chosen based on 65%, 70% and 75% of feature subset. All the feature subsets
results are obtained and the detailed feature subsets are given in the below table 5.
The total features for the given data set is 30 and the subset of features based on
ranking method are shown in below table.
The features are analyzed in each ranking method and union set of all features are
obtained and the results are shown in the table 6 out of 30 features excluding class label.
Initially the performance of the multi-layer perceptron on all the features without
pre-processing are done at the various training split of 66%, 70%,, 72% and 75%
respectively. The performance metrics are given in the table 7.
The performance of a trained classifier based on MLP neural network was
evaluated using four performance measures: correctly classified instances (CCI),
incorrectly classified instances(ICCI), precision, recall, F-Score and ROC. These
measures are defined by four decisions: true positive (TP), true negative (TN),
false positive (FN), and false negative (FN). TP decision occurs when malignant
instances are predicted rightly. TN decision benign instances are predicted rightly.
FP decision occurs when benign instances are predicted as malignant. FN decision
occurs when malignant instances are predicted as benign.

109
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Table 4. Filter Features Selection Attributes

InfoGain
Selected attributes:
23,24,21,28,8,3,4,1,7,14,27,11,13,26,6,17,18,22,2,29,16,25,9,5,30,20,19,10,12,15: 30
GainRatio
Selected attributes:
23,21,24,28,8,7,27,3,4,1,14,6,11,13,26,17,2,19,18,25,22,29,5,16,30,9,20,12,10,15: 30
Chisquare
Selected attributes:
23,21,24,28,8,3,4,1,7,14,27,11,13,26,6,17,18,22,2,29,25,16,9,5,30,20,19,10,12,15: 30
Filtered AttributeEval
Selected attributes:
23,24,21,28,8,3,4,1,7,14,27,11,13,26,6,17,18,22,2,29,16,25,9,5,30,20,19,10,12,15: 30
One R Attribute Evaluator
Selected attributes:
8,21,28,23,3,24,7,1,4,14,27,13,11,6,26,25,22,18,2,29,17,5,30,9,15,16,20,10,12,19: 30
Relief Attribute Evaluator
Selected attributes:
21,28,23,22,1,3,8,24,4,7,2,27,25,11,26,14,10,13,6,5,29,12,19,18,15,30,16,17,9,20: 30
SVM Attribute Evaluator
Selected attributes:
21,28,23,22,8,24,29,1,25,4,11,2,3,7,16,13,10,27,14,9,6,5,15,20,30,12,17,18,19,26: 30
Symmetrical Uncert Attribute Eval
Selected attributes:
23,21,24,28,8,3,7,4,1,27,14,11,13,6,26,17,2,18,22,25,29,16,5,30,9,19,20,10,12,15: 30

Table 5. Filter Features Subset Selection

65% of 70% of 75% of

S.Nos Ranking
Features Features Features
1 InfoGain 20 21 23
2 GainRatio 20 21 23
3 Chisquare 20 21 23
4 Filtered AttributeEval 20 21 23
5 One R Attribute Evaluator 20 21 23
6 Relief Attribute Evaluator 20 21 23
7 Symmetrical Uncert Attribute Eval 20 21 23

110
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Table 6. Final Subset Features

Final Ensemble Features

S.Nos. Ranking Features Subset Total Features Subset
Subset
1 Ensemble Features Subset 1 -65% 20 24
Ensemble Features Subset 2 -70%
2 21 25
Set
3 Ensemble Features Subset 2 -75% 23 26

Table 7. MLP Classifier with all features (30)

S.Nos. Training Split CCI ICCI Precision Recall F-Score ROC

1 Default - 66% 96.37 3.63 0.964 0.964 0.964 0.987
2 70% 97.66 2.34 0.977 0.977 0.977 0.988
3 72% 96.23 3.77 0.962 0.962 0.962 0.987
4 75% 95.77 4.22 0.958 0.958 0.958 0.975

Table 8. MLP Classifier based on 10-Fold Cross Validation with all features (30)

S.Nos. Training Split CCI ICCI Precision Recall F-Score ROC

Cross Validation – All
1 95.66 4.33 0.95 0.95 0.95 0.98
Features
Cross Validation –
2 96.31 3.69 0.96 0.96 0.96 0.99
Features Subset 1
Cross Validation –
3 96.30 3.69 0.96 0.96 0.96 0.99
Features Subset 2
Cross Validation –
4 96.48 3.51 0.97 0.97 0.97 0.99
Features Subset 3
Training Split – 70% –
5 97.07 2.92 0.97 0.97 0.97 0.98
Features Subset 3

To evaluate the performance of above methods of neural network training

different parameters are accuracy, precision, recall, F-Measure, kappa score etc.
For our proposed system the metrics used are accuracy, precision, recall and ROC
are considered. Accuracy gives a measured value in par with the actual (true) value
and retrieves the percentage of correctly classified instances.

111
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

6. CONCLUSION

In this paper, an automated computer aided diagnosis system has been devised
for diagnosis of breast cancer. The performance are measured and computed with
supervised neural network classifier model. The produced classification results are
very much promising with 97% accuracy of correct classification and F-Score is of
93% with reduced error measures is of 4% by including all features. The proposed
algorithm gives up an ensemble feature subsets and this is applied to artificial neural
network with a better accuracy for breast cancer diagnosis. The proposed method
may provide an adequate support to the radiologists in differentiating between
normal and abnormal breast cancer identification with high accuracy and of low
error measures. The research can be focused further to develop better preprocessing,
enhancement and segmentation techniques. The proposed work can be further
expanded to design enhanced feature extraction and selection and also appropriate
classification algorithms can be used to reduce both false positives and false negatives
by employing high resolution mammograms and investigating 3D mammograms.

REFERENCES

Arafi, A., Safi, Y., Fajr, R., & Bouroumi, A. (2013). Classification of Mammographic
Images using Artificial Neural Network. Applied Mathematical Sciences, 7(89),
4415–4423. doi:10.12988/ams.2013.35293
Chang, R. F., Wu, W. J., Moon, W. K., & Chen, D.-R. (2005). Dr Chen, “Automatic
Ultrasound Segmentation and Morphology based Diagnosis of Solid Breast Tumors.
Breast Cancer Research and Treatment, 89(2), 179–185. doi:10.100710549-004-
2043-z PMID:15692761
Chen, Y., Wang, Y., & Yang, B. (2006). Evolving Hierarchical RBF Neural Networks
for Breast Cancer Detection. LNCS, 4234, 137-144. doi:10.1007/11893295_16
Hazlina, H., & Sameem, A. K. (2004). Back Propagation Neural Network for
the Prognosis of Breast Cancer: Comparison on Different Training Algorithms.
Proceedings Second International Conference on Artificial Intelligence in
Engineering & Technology, 445-449.
Horsch, K., Giger, M. L., Venkata, L. A., & Vybomya, C. J. (2001). Automatic
Segmentation of Breast Lesions on Ultrasound. Medical Physics, 28(8), 1652–1659.
doi:10.1118/1.1386426 PMID:11548934

112
An Ensemble Feature Subset Selection for Women Breast Cancer Classification

Hotko, Y. S. (2013). Male breast cancer: Clinical presentation, diagnosis, treatment.

Experimental Oncology, 35(4), 303–310. PMID:24382442
Kiyan, T., & Yildrim, T. (2004). Breast Cancer Diagnosis using Statistical Neural
Networks. Journal of Electrical and Electronics Engineering (Oradea), 4(2),
1149–1153.
Madabhushi, A., & Metaxas, D. (2003). Combining low-, high-level and Empirical
Domain Knowledge for Automated Segmentation of Ultrasonic Breast Lesions. IEEE
Transactions on Medical Imaging, 22(2), 155–169. doi:10.1109/TMI.2002.808364
PMID:12715992
Malvia, S., Bagadi, S. A., Dubey, U. S., & Saxena, S. (2017). Epidemiology ofbreast
cancer in Indian women. Asia Pacific Journal of Clinical Oncology, 13(4), 289–295.
doi:10.1111/ajco.12661 PMID:28181405
Raad, A., Kalakech, A., & Ayache, M. (2012). Breast Cancer Classification using
Neural Network Approach: MLP AND RBF. The 13th international Arab conference
on information technology, 10 – 13.
Sahiner, B., Heang-Ping, C., Patrick, N., Wei, D. M. A., Helie, D., Adler, D., &
Goodsitt, M. M. (1996). Classification of Mass and Normal Breast Tissue: A
Convolution Neural Network Classifier with Spatial Domain and Texture Images.
IEEE Transactions on Medical Imaging, 15(5), 598–610. doi:10.1109/42.538937
PMID:18215941
Sapra, P., Singh, R., & Khurana, S. (2013). Brain Tumor Detection using Neural
Network. International Journal of Science and Modern Engineering, 1(9).
Shallu, R. M., & Mehra, R. (2018). Breast cancer histology images
classification:Training from scratch or transfer learning? ICT Express, 4(4), 247–254.
doi:10.1016/j.icte.2018.10.007
Singh, S., Saini, S., & Singh, M. (2012). Cancer Detection using Adaptive Neural
Network. International Journal of Advancements in Research and Technology, 1(4).
Wadhwani & Saraswat. (2009). Classification of breast cancer using artificial neural
network. Current Research in Engineering. Science and Technology Journals.
Wu, Y., Wang, N., Zhang, H., Qin, L., Yan, Z., & Wu, Y. (2010). Application of
Neural Networks in the Diagnosis of Lung Cancer by Computed Tomography. Sixth
International Conference on Natural Computation. 10.1109/ICNC.2010.5583316

113
114

Chapter 7
A Content-Based Approach
to Medical Image Retrieval
Anitha K.
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

Naresh K.
VIT University, India

Rukmani Devi D.
RMD Engineering College, India

ABSTRACT
Medical images stored in distributed and centralized servers are referred to for
knowledge, teaching, information, and diagnosis. Content-based image retrieval
(CBIR) is used to locate images in vast databases. Images are indexed and retrieved
with a set of features. The CBIR model on receipt of query extracts same set of
features of query, matches with indexed features index, and retrieves similar images
from database. Thus, the system performance mainly depends on the features
adopted for indexing. Features selected must require lesser storage, retrieval time,
cost of retrieval model, and must support different classifier algorithms. Feature
set adopted should support to improve the performance of the system. The chapter
briefs on the strength of local binary patterns (LBP) and its variants for indexing
medical images. Efficacy of the LBP is verified using medical images from OASIS.
The results presented in the chapter are obtained by direct method without the aid
of any classification techniques like SVM, neural networks, etc. The results prove
good prospects of LBP and its variants.

DOI: 10.4018/978-1-7998-3092-4.ch007

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
A Content-Based Approach to Medical Image Retrieval

INTRODUCTION

Due to the enormous size of medical image data repository, CBIR can be used for
medical image retrieval. This chapter is envisioned to propagate the knowledge of
the CBIR approach to deal with the applications of medical image management
and to pull in more prominent enthusiasm from various research groups to rapidly
propel research in this field.
The image is presumably a standout amongst the most essential tools in medicine
since it provides a method for diagnosis, monitoring drug treatment responses and
disease management of patients with the advantage of being a very fast non-invasive
procedure, having very few side effects and with an excellent cost-effect relationship.
Hard-copy image formats used to support for medical images are not utilized
these days. The expense and resource involved in maintenance, storage room and
the amount of material to display images in this format contributed for its disuse.
Nowadays digital images, that doesn’t face problems mentioned for hard copy
formats are used. Table 1 gives a review of digital images per exam in medical
imaging. This transition from hard-copy to soft-copy images is still the center of an
interesting debate related with human perception and understanding issues during
exam analysis. Elizabeth (2000) have tended to the significance of observation in
medical imaging.

Table 1. Types and sizes of some commonly used digital medical images from Huang
(2004)

No. of Images/ One

Image Type One Image(bits)
Exam Examination
Nuclear medicine (NM) 128X128X12 30-60 1-2 MB
Magnetic resonance imaging (MRI) 256X256X12 60-3000 8 MB
Ultrasound (US)* 512X512X8 20-240 5-60 MB
Digital subtraction angiography (DS) 512X512X8 15-40 4-10 MB
Digital microscopy 512X512X8 1 0.25 MB
Digital color microscopy 512X512X24 1 0.75 MB
Color light images 512X512X24 4-20 3-15 MB
Computed tomography (CT) 512X512X24 40-3000 20 MB
Computed/digital radiography (CR/DR) 2048X2048X12 2 16 MB
Digitized X-rays 2048X2048X12 2 16 MB
Digital mammography 4000X5000X12 4 160 MB
*Doppler US with 24 bit color images

115
A Content-Based Approach to Medical Image Retrieval

Increase of medicinal information in digital libraries makes tougher to perform

analysis on search related tasks. Since textual information retrieval is as of now a
developed discipline, an approach to overcome this issue is to utilize metadata for
image indexing. Key description, patient identification, kind of exam and its technical
details or even a small text comment concerning clinical relevant information can
be utilized to represent the image in its index. With these annotations, text-matching
techniques that assess the similarity between the search statement and the metadata
can be applied for retrieving images. This is called text-based or concept-based
image retrieval.

LIMITATIONS OF CONCEPT-BASED
RETRIEVAL AND THE NEED FOR CBIR

Generally to archive an image with a general vocabulary of medical terms devours

numerous resources and requests broad collaboration efforts that is difficult to
incorporate. It is sensible to utilize inductive methodologies by beginning with more
particular standards and endeavor generalization later. This strategy is utilized by (Dean
Bidgood 1998) in the composite SNOMED-DICOM micro-glossary. Annotation
of images by human is a time consuming and cumbersome task and also lead to
unrecoverable errors. (Mark et al 2002) has undergone the study of medical images
using DICOM headers and revealed that 15% of annotation errors occur from both
human and machine origin. Another significant obstacle in concept-based image
retrieval systems is the fact that the query, does not allow the user to switch and/or
combine interaction paradigms (Shi-kuo and Arding 1992) during text transactions.
The ideal framework would relieve the human from the annotation task, by doing
it automatically, and allowing image retrieval by its content instead of textual
description. This framework is known as Content Based Image Retrieval (CBIR).

CBIR IN MEDICAL APPLICATIONS

The CBIR in the medical field presents a developing trend in publications. Comparative
analysis of CBIR implementations in medical imaging are presented by (Long et al
2009 and Hussain et al 2020). A study proposed by (Ogul et al 2020) implemented
a new method of PD detection from gait signals, using artificial neural networks
and a novel technique framework called Neighborhood Representation Local Binary
Pattern (NR-LBP). Vertical Ground Reaction Force (VGRF) readings are preprocessed
and transformed using several methods within the proposed framework. Despite
the growth of CBIR frameworks in medical imaging, the utility of the frameworks

116
A Content-Based Approach to Medical Image Retrieval

in number of medical applications is still very constrained. There exist only a

couple of systems with relative success. The Cervigram Finder system (Zhiyun et
al 2008) was developed to study the uterine cervix cancer. It is a computer assisted
framework where local features from a user-defined region in an image are computed
and, using similarity measures, similar images are retrieved from a database. The
Image Retrieval for Medical Applications (IRMA) framework (Thomas et al 2008)
is a web-based x-ray retrieval system. It permits the user to retrieve images from
a database given an x-ray image as query. CBIR for medical applications can be
found in (Lehmann et al 2004 and Muller et al 2007). Medical applications are one
of the priority areas where CBIR can meet more accomplishment due to population
aging in developed countries.

PROBLEM DEFINITION

Generally to archive an image with a general vocabulary of medical terms devours

numerous resources and requests broad collaboration efforts that is difficult to
incorporate. It is sensible to utilize inductive methodologies by beginning with more
particular standards and endeavor generalization later. Another significant obstacle
in concept-based image retrieval systems is the fact that the query, does not allow
the user to switch and/or combine interaction paradigms during text transactions.
The ideal framework would relieve the human from the annotation task, by doing it
automatically, and allowing image retrieval by its content instead of textual description.
This framework is known as retrieving images with the help of the image content.
Increase of medicinal information in digital libraries and availability of advanced
technologies to store and manage shall be utilized effectively in diagnosis and treating
patients. Searching and retrieving similar cases from the huge database plays a vital
role in developing an efficient architecture to perform the task. The information stored
shall be in the form of medical images and text records. Medical Image Retrieval
Systems (MIRS) with determined feature vector to index the images and adopting
advanced Classifying methods shall be more efficient to cluster and locate similar
medical images or records stored in vast databases.
The proposed work concentrates on classification of medical images provided
by OASIS. The work aims at the following goals:

• Implement and analysis a simple, effective and efficient medical image

retrieval system
• To use Local Binary Pattern (LBP operator) to represent medical images and
develop a retrieval system that match and retrieve images with the support of
the operator.

117
A Content-Based Approach to Medical Image Retrieval

• To derive hybrid descriptors combining two LBP operators with different

sizes of neighborhood pixels and study the impact on the retrieval results
• To point new directions and considerations where future work can be
developed

IMAGE DOMAIN

The digital image is an approximation of a two-dimensional image by set of values

called pixels. Each pixel is described in terms of its color, intensity/luminance or
value. In order to retrieve images according to a given query we need to enhance
its relevant elements while reducing the remaining aspects. This is the goal of image
processing. Generically, the image is processed using an operator, g, over the full
spatial domain of the image, I (X ,Y ) , an interest point, I (x n , ym ) to generate a
feature space containing the information needed to identify the objects in the following
way:

f (r ) = g I (x , y ) (1)

f (r(x ,y ) ) = g I (x , y ) (2)

f (r(x ,ym )
) = g I (x n , ym ) (3)
n

Where I (X ,Y ) is the full image; I (x , y ) is an image patch, i.e., a connected subset

of Cartesian points with (x , y ) ∈ (X ,Y ), ∀x , y ∈ R and I (x n , ym ) is at interest point
(x n , ym ) where n ∈ {1, 2, 3,....., N x } and m ∈ {1, 2, 3,....., M x } .

Image Properties

The relationships between image properties like color, shape, texture and interest
points are the fundamental characteristics of an image. Thus the properties used to
represent images mathematically are named as image descriptors. The similarity
between images is also estimated using these descriptors.

118
A Content-Based Approach to Medical Image Retrieval

Image Descriptors

It is very difficult to ascertain which image properties are fundamental to characterize

a specific image. It depends on the context of the problem to be solved and the
knowledge within the image itself. If there is a need to recognize specific objects
in a scene, probably the shape property is more relevant than the others. However
if such objects have a distinct color then the relevance of this property is higher
than the rest. If there is a need to detect light bulbs in a night scenario we rely on an
interest point detector. Sometimes the image can be quite complex and all properties
are essential for its characterization. Eq. (1) to (3) enables the quantification of the
image properties. This quantification bridges the gap between human perception
of image properties and mathematical measure(s) taken from the image. This is the
aim of an image descriptor.
Image descriptors can be global, like Eq. (1), or local, like Eq. (2) and Eq. (3).
The feature database is constructed using global descriptor that represents the visual
features of the whole image as well as the local descriptor that represents the visual
features of regions or objects to describe the image. Similarity measurement is done
by many methods like Euclidean distance (L2), L1 distance etc. The selection of
feature descriptors and similarity metrics greatly affects the retrieval performance.
The description of the features available in the literature for biomedical imaging is
briefed as follows. (Hersh et al 2009) proposed Image CLEF medical image retrieval
task to improve understanding and system capability in search for medical images.
They described the development and use of a medical image test collection design
to facilitate research with image retrieval systems and their users. (Manjunath et al
2007) presented bit plane histogram and hierarchical bit plane histogram along with
cumulative distribution function (CDF) for CT and MRI image retrieval. (Fahimeh
et al 2010) has done the classification of benign and malignant breast masses based
on shape and texture features in sonography images. The blood cell image retrieval
using color histogram and wavelet transform can be seen in the work proposed by
(Woo and Seyed 2009).
Medical image retrieval system detailed in this chapter uses Local binary patterns
(LBP) proposed by (Ojala et al 1996) to represent images in database. LBP can show
better performance as well as less computational complexity for image classification.
LBP have been used in many research areas such as texture classification (Guo et
al 2010, Xueming et al 2011, Loris et al 2012), face recognition (Zhang et al 2010,
Tan et al 2010, Papakostas et al 2013), and image retrieval(Cheng-Hao et al 2003,
Sorensen et al 2010).

119
A Content-Based Approach to Medical Image Retrieval

LBP METHODOLOGY

The image descriptors developed for representing images mathematically shows more
emphasis on effectiveness and less on efficiency. Effectiveness is the closeness of
the retrieved images to the query. Efficiency is the utilization of optimum resources
to retrieve more similar images from the datasets. The need is an effective and
efficient system.
Focuses of the chapter is to brief an analysis and implementation of an image
descriptor to retrieve medical images that speed up the process by reducing the
dimension of image descriptor. To get real time processing speeds, less dimension
of the image descriptor is an important property of the methods which take less
time to build the feature and less time to retrieve the matched images. However,
lower dimension usually has less distinctiveness than higher dimension. Therefore,
we must need to take care of this tradeoff while building the descriptor to represent
the image. Among the image description methods, local binary pattern (LBP) has
received considerable attentions in many computer vision applications, such as
face recognition, image retrieval and motion analysis, for its efficiency and simple
computational complexity to build the image descriptor. LBP method had been
proposed for texture analysis. This operator is defined as a monotonic illumination
invariant texture measure, derived from local neighborhood. For each pixel in an
image, a binary code is produced by thresholding center intensity value with the
intensity value of the neighbor pixel. Then, A histogram, created to collect the
occurrences of different binary patterns, is used to represent the image. The basic
LBP operator considers only the eight neighbors of a center pixel. However, the
definition has been extended to include any number of circular neighborhoods by
using the interpolation technique
The LBP operator was initially presented as a complementary measure for local
image contrast (Ojala et al 1996). An approach that opted to adopt Local Binary
Patterns and k-means clustering for precise identification of lesions boundaries
(Pedro et al 2020), particularly the melanocytic has been derived. A blind detection
based uniform local binary patterns (ULBP) is proposed (Zhang et al 2020) to detect
seam-carved image. The gentle boost decision trees are trained(Gogic et al 2020) to
extract highly discriminative feature vectors (local binary features) for each basic
facial expression around distinct facial landmark points for faster fascial recognition.
The operator works with the eight-neighbors of a pixel, using the center pixel value
as a threshold. LBP code for a neighborhood was generated by multiplying the
threshold values with weights assigned to the corresponding pixels, and summing
up the result Figure 6.1. Since the LBP was, invariant to monotonic changes in gray
scale, it was supplemented by an orthogonal measure of local contrast.

120
A Content-Based Approach to Medical Image Retrieval

Figure 1. Basic Version of LBP operator

Derivation

The texture T in a local neighborhood of a grayscale image is defined as the joint

distribution of the gray levels of P + 1 (P > 0) image pixels:

T = t(gc , g 0 ......., g p−1 ) (4)

where gc corresponds to the gray value of the center pixel of a local neighborhood.
g p (p = 0, 1, 2, 3......, P − 1) correspond to the gray values of P pixels equally spaced
on a circle of radius R (R > 0) that form a circularly symmetric set of neighbors.
This set of P + 1 pixels is later denoted byGP . In a digital image domain, the
coordinates of the g P neighbors are given by

(x c + R cos(2Πp / P ), yc − R sin(2Πp / P ))

where (xc , yc ) are the coordinates of the center pixel. Figure 2 illustrates three
circularly symmetric neighbor sets for different values of P and R.

Figure 2. Circularly symmetric neighbor sets. Samples that do not exactly match
the pixel grid are obtained via interpolation.

121
A Content-Based Approach to Medical Image Retrieval

The values of neighbors that do not fall exactly on pixels are estimated by bilinear
interpolation and the operator can be expressed using

T = t(gc , g 0 − gc ......., g p−1 − gc ) (5)

Assuming that the differences are independent of gc , the distribution can be

factorized:

T ≈ t(gc ), t(g 0 − gc ......., g p−1 − gc ) (6)

Since t(gc ) describes the overall luminance of an image, which is unrelated to

local image texture, it does not provide useful information for texture analysis Eq.
6 simplifies to

T ≈ t(g 0 − gc ......., g p−1 − gc ) (7)

The P−dimensional difference distribution records the occurrences of different

texture patterns in the neighborhood of each pixel. For constant or slowly varying
regions, the differences cluster near zero. On a spot, all differences are relatively
large. On an edge, differences in some directions are larger than the others. Although
invariant against gray scale shifts, the differences are affected by scaling. To achieve
invariance with respect to any monotonic transformation of the gray scale, only the
signs of the differences are considered:

T ≈ t(s(g 0 − gc )......., s(g p−1 − gc )) (8)

where

1 s ≥0
s(x ) = {0 s <0
(9)

Now, a binomial weight 2p is assigned to each signs(g p − gc ) , transforming the

differences in a neighborhood into a unique LBP code. The code characterizes the
local image texture around(x c , yc )

122
A Content-Based Approach to Medical Image Retrieval

P −1
LBPP ,R (x c , yc ) = ∑ s(g p − gc )2p (10)
p

In practice, Eq. 6.8 means that the signs of the differences in a neighborhood are
interpreted as a P-bit binary number, resulting in 2P distinct values for the LBP code.
The local gray-scale distribution, i.e. texture, can thus be approximately described
with a 2P bin discrete distribution of LBP codes:

T ≈ t(LBPP ,R (xc , yc )) (11)

Let us assume we are given an NXM image sample

(x c ∈ {0, 1, 2,...., N − 1}, yc ∈ {0, 1, 2...., M − 1}) .

In calculating the LBPP ,R distribution (feature vector) for this image, the central
part is only considered because a sufficiently large neighborhood cannot be used
on the borders. The LBP code is calculated for each pixel in the cropped portion of
the image, and the distribution of the codes is used as a feature vector, denoted by
S

S = t(LBRP ,R (x , y )) (12)

where

x ∈ {[R ],......, N − 1 − [R ]}, y ∈ {[R ],......, M − 1 − [R ]} .

Combination of the Structural and

Stochastic Approaches by LBP

The LBP operator to represent images is a unifying approach and can be considered
as a micro-texton. The texture primitive that best matches the nearby neighborhood is
used to label every pixel. Spots, edges, curves, flat areas, edge ends etc. are detected
by local primitives of LBP.
Figure 3 provides few of such illustrations. White and black circles in the figure,
represents ones and zeros respectively.

123
A Content-Based Approach to Medical Image Retrieval

Figure 3. Different texture primitives detected by the LBP

Statistical and structural methods have been normally used separately to detect
the textures. The LBP technique has both of these properties, texture primitives and
placement rules. Thus the operator is a better option to distinguish/match a variety
of texture images

Rotation Invariance

Circular sampling of neighborhoods to form the operator supports to make LBP

operator invariant to the rotation of an image. The below assumptions are considered
to derive the operator.

• Rotation invariance here does not count for textural differences caused by
changes in the relative positions of a light source and the target object.
• The effects caused due to digitizing effects are neglected.
• Each pixel is considered a rotation center

The gray values g p in a neighbor set of the circle move along the perimeter
centered at gc due to the rotation of the image. LBPP ,R Value will be different for a
rotated image from that of the actual. Patterns comprising of only zeros or ones will
remain unchanged with rotation. Rotating the LBP operator back to a reference
position will eliminate the effect of rotation and makes versions of a binary code
the same. This transformation can be defined as follows:

LBPPri,R = min{ROR(LBPP ,R , i )∀i = 0, 1,...., P − 1} (13)

where the superscript ri stands for rotation invariant. The function ROR(x , i )
circularly shifts the P-bit binary number x by i times to the right ( i < p) . That
is, given a binary number x :

124
A Content-Based Approach to Medical Image Retrieval

P −1
x = ∑ 2k ak ∀ak = {0, 1} (14)
k =0

the ROR operation is defined as:

 P −1 2k −ia + i −1

∑ k =i k ∑ k =0 2P −i +kak i > 0

ROR(x , i ) = x i = 0 (15)

ROR(x , P + i ) i < 0


In short, the rotation invariant code is produced by circularly rotating the original
code until its minimum value is attained. Figure 6.4 illustrates six rotation invariant
codes in the top row. Below these, examples of rotated neighborhoods that result in
the same rotation invariant code are shown. In total, there are 36 different 8-bit
rotation invariant codes. Therefore, LBR8,riR , produces 36-bin histograms.

Figure 4. Neighborhoods rotated to their minimum value (top row) and that produce
the same rotation invariant LBP codes.

The LBP codes shown in Figure 3 are all uniform. Examples of non-uniform
codes can be seen in Figure 4, in the third and fifth columns. To formally define the
uniformity of a neighborhood G, a uniformity measure U is needed:

P −1
U (G ) = s(g P −1 − g c ) − s(g 0 − g c ) + ∑ s(g p − g c ) − s(g p−1 − g c ) (16)
p =1

125
A Content-Based Approach to Medical Image Retrieval

Patterns with a U value of less than or equal to two are designated as uniform.
For a P-bit binary number, the U value can be calculated efficiently as follows:

P −1
U (x ) = ∑ Fb (x xor ROR(x , 1), p) (17)
p =0

where b is a binary number. The function Fb (x , i ) extracts b circularly successive

bits from a binary number x

Fb (x , i ) = ROR(x , i )and (2b − 1) (18)

where i is the index of the least significant bit of the bit sequence. The operators
“and” and “xor” denote bitwise logical operations.
The total number of patterns with U(GP)≤2 is P(P−1)+2. When uniform codes
are rotated to their minimum values, the total number of patterns becomes P +
1. The rotation invariant uniform (riu2) pattern code for any uniform pattern is
calculated by simply counting ones in the binary number. All other patterns are
labeled “miscellaneous” and collapsed into one value

 P −1
= ∑ p =0 p
 s(g − gc )U (GP ) < 2
LBP riu 2
(19)
P ,R
P + 1otherwise


In practice, LBPPriu,R2 is best implemented by creating a look-up table that converts

the “basic” LBP codes into their LBPPriu,R2 correspondents.

Contrast and Texture Patterns

The LBP operator ignores the amount of gray level divergences. But the magnitude
of gray level that provides the contrast is a property of texture and is important for
our vision system to arrive a result. An operator that is not influenced by gray scale
may waste useful information obtained from applications that have a reasonable
control on illumination accurately. The accuracy of the operator can be enhanced
by including information about gray-scale. Texture is identified with two properties
spatial structure and contrast. Spatial structure is independent on gray scale and
affected by rotation whereas contrast is dependent on gray scale but not affected by
rotation. A joint distribution of LBP operator and local contrast measure (LBP/C)
as a texture descriptor has been implemented by (Ojala et al 1996).

126
A Content-Based Approach to Medical Image Retrieval

VARP ,R does not change due to variations in the gray scale. It can be estimated
in circular sets as similar to LBP.

1 P −1
VARP ,R = ∑
P p =0
(g p − µ)2
(20)
1 P −1
where µ = ∑ g p
P p =0

An operator denoted as LBPPriu1,R21 /VARP 2,R 2 formed with joint distribution of

LBP and local variance can be formed. The texture descriptor thus formed will not
vary with rotation.

EXPERIMENTAL RESULTS

OASIS (Open Access Series of Imaging Studies) is a series of magnetic resonance

imaging (MRI) dataset that is publicly available for study and evaluation. This
dataset proposed by (Marcus et al 2007) comprises cross-sectional collection of 421
subjects. The images are grouped into four categories (Group-1: 124, Group-2:102,
Group-3: 89, and Group-4: 106 images) and used for experimentation of proposed
CBIR system. Grouping is done based on the shape of ventricular in the images.
For training the classifier 30% of images in each category are utilized and 30% and
40% of images are used for validation and testing respectively.
The purpose of experiments is to estimate and prove the capability of the LBP
operator to represent the medical images mathematically. Experiments are performed
separately to test the capacity of the operator to classify the images and to retrieve
the images. A basic retrieval system that classifies and retrieves images using LBP
(without help of trained classifier) is tested and the results are shown.

Similarity Measure

The objective of any CBIR system is to archive n best images from an image database
(N number of images) that resemble the query image. The selection of n images
that best matches is selected by measuring the distance between query image and
N images in the database. In literature it is found four types of similarity distance
metrics have been used for the purpose.
L1 or Manhattan distance measure

127
A Content-Based Approach to Medical Image Retrieval

lf
D(Q, DB j ) = ∑ ( fDBji − fQi ) (21)
i =1

L2 Euclidean distance measure:

lf 2

D(Q, DB j ) = ∑ (fi =1
DBji
− fQi ) (22)

Canberra distance measure:

lf (f − fQi )
D(Q, DB j ) = ∑
DBji
(23)
i =1 fDBji + fQi

d1 distance measure:

D(Q, DB j ) = ∑
lf
(f DBji
− fQi )
(24)
i =1 1 + fDBji + fQi

Where fDBji is the jth (length of feature vector is lf ) feature of ith image in Data base
DB of N images.

Evaluation Metrics

The performance of the proposed system is evaluated by the parameters average

retrieval precision (ARP) and average retrieval rate (ARR). The evaluation parameters
are calculated using Eqs. (26)–(29). The equations also define the precision (P) and
recall (R) for the query image Iq.

Number of relevant images retrieved

Precision P (I q ) = (25)
Number of images retrieved

N
1
ARP =
N
∑ P(I )
i =1
i
(26)
n ≤10

128
A Content-Based Approach to Medical Image Retrieval

Number of relevant images retrieved

Recall R(I q ) = (27)
Total number of relevant images in data base

N
1
ARR =
N
∑ R(I )
i =1
i
(28)
n ≥10

In order to analyze the performance of our algorithm biomedical image retrieval

is performed on two different medical databases. Results obtained are discussed in
the following subsections. In all experiments, each image in the database is used
as the query image. For each query, the system collects n database images with the
shortest image matching distance is given by Eq. (26-28). If the retrieved image
belongs to the same category of the query image, we say the system has correctly
matched the desired.

Classification and Retrieval Using Basic Retrieval System

Results in Table 2 illustrate the group wise percentage of correctly retrieved images
from OASIS-MRI database. In this chapter, experiments are performed with individual
LBP and VAR features of images and by combination of both to develop a hybrid
system. In addition to the improvements achieved with hybrid features, the combined
features significantly found robust under illumination variations and invariance to
rotation. It has been observed from detailed experimental results that LBP16riu,22 and
VAR8,1 supersede other variants considered individually or if combined for various
images groups. The accuracy of the retrieval system is improved further by integrating
variants of LBP and VAR. Several experiments are performed to analyze and confirm
the superiority of the proposed hybrid approach. LBP16riu,22 /VAR8,1 based hybrid
approach also improves the image retrieval rate but slightly less than LBP8riu
,1
2
/ LBP24riu,32 .
It has also been observed that hybrid approach is more effective for extracting
features offline and for online retrieval in comparison to other hybrid methods.
Therefore, all the experiments performed in this chapter lead to proving LBP based
indexing, the proposed hybrid approach is more superior.
From Table 2, the following inference is drawn. LBP16riu,22 and LBP24riu,32 clearly
outperformed their simpler counterpart LBP8riu ,1
2
, which had difficulties in
discriminating strongly oriented texture. In nearly 176 test cases, the system using
LBP8riu
,1
2
identified true class of 74 test samples, LBP16riu,22 identified 79 test samples
LBP24riu,32 identified 78 test samples correctly. If group wise classification of images

129
A Content-Based Approach to Medical Image Retrieval

is considered, LBP16riu,22 did much better, classifying more than 45% samples correctly.
Combining the LBPPriu,R2 operator with the VARP ,R operator, improved the
performance. It was observed that LBP16riu,22 /VAR16,2 provided more comparable
results. It is noticed from the classification results the combined features of LBP
and VAR aids to improve the search and retrieval process. From the evaluations it
is evident that LBP and VAR are redundant and provide excellent results under
illumination variations and texture images

Table 2. Group wise percentage of correctly retrieved images from OASIS-MRI

database

Category
Method
1 2 3 4 Avg

LBP8riu
,1
2
54.10 36.50 38.89 41.86 42.04

LBP16riu,22 58.00 39.02 36.11 55.81 46.59

LBP24riu,32 48.00 46.34 38.89 48.83 44.31

VAR8,1 58.00 41.46 41.66 44.18 45.45

VAR16,2 54.00 36.58 36.11 37.20 40.34

VAR24,3 52.50 36.58 38.88 34.88 39.77

LBP8riu
,1
2 riu 2
/ LBP16,2 68.00 53.66 58.33 53.49 56.82

LBP8riu
,1
2 riu 2
/ LBP24,3 72.00 63.41 63.89 55.14 61.93

LBP16riu,22 / LBP24riu,32 68.00 60.98 61.11 51.16 58.52

VAR8,1 /VAR16,2 62.00 46.34 47.22 48.84 50.00

VAR8,1 /VAR24,3 54.00 43.90 44.44 46.51 46.02

VAR16,2 /VAR24,3 62.00 58.54 58.33 51.16 55.68

continues on following page

130
A Content-Based Approach to Medical Image Retrieval

Table 2. Continued

Category
Method
1 2 3 4 Avg

LBP8riu
,1
2
/VAR8,1 55.35 48.78 52.776 48.83 51.70

LBP8riu
,1
2
/VAR16,2 62.00 51.21 50.00 44.18 50.00

LBP8riu
,1
2
/VAR24,3 60.00 48.78 55.00 44.18 49.43

LBP16riu,22 /VAR8,1 70.00 63.41 63.89 51.16 60.23

LBP16riu,22 /VAR16,2 64.00 51.21 58.33 51.16 54.44

LBP16riu,22 /VAR24,3 58.00 48.78 47.22 44.18 48.86

LBP24riu,32 VAR8,1 68.00 58.54 63.89 48.84 57.95

LBP24riu,32 VAR16,2 62.00 43.90 44.44 55.81 50.57

LBP24riu,32 VAR24,3 60.00 48.78 52.77 48.83 51.13

Table 3. Percentage of classification accuracy achieved using LBP operator with

different distance measures on OASIS database

Distance Measure
Method
L1 L2 Canberra d1

LBP8riu
,1
2 riu 2
/ LBP16,2 56.82 55.11 57.39 59.09

LBP8riu
,1
2 riu 2
/ LBP24,3 61.93 59.09 63.64 65.91

LBP16riu,22 / LBP24riu,32 58.52 55.68 59.66 61.36

LBP16riu,22 /VAR8,1 60.23 57.95 61.93 63.64

LBP24riu,32 /VAR8,1 57.95 56.25 60.80 62.50

131
A Content-Based Approach to Medical Image Retrieval

Table 3 illustrates the percentage of classification accuracy achieved using LBP

operator with different distance measures. It is clear that d1 similarity measure shows
better performance as compared to other measures.
Figure 5 shows the retrieval performance of CBIR system with LBP/VAR operators
in terms of ARP and ARR as a function of number of top matches and Figure 6 and
7 provides the snap shots of MIRS. The system performs better, if the required number
of similar images are lesser. That is if the database has 25 images similar to query
image and the retrieval system is requested for 10 similar images, then the system
can retrieve 8-10 similar images. In the other hand if the system is asked for retrieving
25 images, then 13-15 images are retrieved. The system performance declines with
increase in requirement. The system performance is little less consistent. The same
can be improved by combining global features with LBP. Since our interest is to
evaluate the performance of LBP features, the investigations of combining LBP with
global features shall be considered as a separate study. The retrieval performance of
hybrid approaches is determined in terms of precision and recall (ARP − ARR)
curves. The comparison is performed in terms of ARP − ARR curves. The analysis
is performed for OASIS-MRI dataset in which images of the same class show various
variations. It is observed that among the hybrid LBP features, the superior performance
is given by LBP8riu ,1
2
/ LBP24riu,32 . While observing the performance of combined LBP,
we see that the proposed approach effectively overpowers other local and achieves
a very high retrieval rate. It is apparent from the ARP − ARR curves.

SUMMARY

The work presented a theoretically and computationally simple multi resolution

approach to gray scale and rotation invariant texture classification. A generalized
gray scale and rotation invariant operator LBPPriu,R2 for detecting ‘uniform’ patterns
in circular neighborhoods is employed to represent images. To experiment the
rotation invariance, the classifier was trained at one particular rotation angle and
tested with samples from other rotation angles.
Retrieval results proved the proposed approach is very robust in terms of gray
scale variations caused e.g. by changes in illumination intensity, since the LBPPriu,R2
operator is by definition invariant against any monotonic transformation of the gray
scale. This should make it very attractive in situations where non uniform illumination
conditions are a concern, e.g. in visual inspection. Gray scale invariance is also
necessary if the gray scale properties of the training and testing data are different.
The performance can be further enhanced by multi resolution analysis, combining
Global features such as WM, ART, PCTs etc. with LB and adopting advanced
classification techniques involving Machine learning algorithms.
132
A Content-Based Approach to Medical Image Retrieval

Figure 5. Performance of CBIR system with LBP/VAR operators in terms of ARP

and ARR as a function of number of top matches

Figure 6. Medical Image Retrieval System with LBP/VAR operators for images from
OASIS datasets

133
A Content-Based Approach to Medical Image Retrieval

Figure 7. Snap shots for retrieval results with different classifiers on database of
human parts

REFERENCES

Chang, S.-K., & Hsu, A. (1992). Image information systems: Where do we go from
here? IEEE Transactions on Knowledge and Data Engineering, 4(5), 431–442.
doi:10.1109/69.166986
Dean Bidgood, W. Jr. (1998). The SNOMED DICOM Microglossary: Controlled
terminology resource for Idata interchange in biomedical imaging. Methods
of Information in Medicine, 37(4/5), 404–414. doi:10.1055-0038-1634557
PMID:9865038
Gogic, I., Manhart, M., Pandzic, I. S., & Ahlberg, J. (2020). Fast facial expression
recognition using local binary features and shallow neural networks. The Visual
Computer, 36(1), 97–112. doi:10.100700371-018-1585-8
Guld, Kohnen, Keysers, Schubert, Wein, Bredno, & Lehmann. (2002). Quality of
DICOM header information for image categorization. SER, Proc. SPIE, 4685, 280-
287. 10.1117/12.467017
Guo, Z., Zhang, L., & Zhang, D. (2010). Rotation invariant texture classification
using LBP variance with global matching. Pattern Recognition, 43(3), 706–716.
doi:10.1016/j.patcog.2009.08.017
Hersh, W., Muller, H., & Kalpathy. (2009). The imageCLEFmed medical image
retrieval task test collection. J. Digital Imaging, 22(6), 648-655.
Huang, H. K. (2004). PACS and imaging informatics: basic principles and applications.
John Wiley & Sons Inc. doi:10.1002/0471654787
Hussain, C. A., Rao, D. V., & Mastani, S. A. (2020). RetrieveNet: a novel deep
network for medical image retrieval. Evol. Intel. doi:10.100712065-020-00401-z

134
A Content-Based Approach to Medical Image Retrieval

Krupinski. (2000). The Importance of Perception Research in Medical Imaging.

Radiation Medicine, 8(6), 329-334.
Lehmann, T. M., Guld, M. O., Thies, C., Fischer, B., Spitzer, K., Keysers, D., Ney,
H., Kohnen, M., Schubert, H., & Wein, B. B. (2004). Content-based image retrieval
in medical applications. Methods of Information in Medicine, 43(4), 354–361.
doi:10.1055-0038-1633877 PMID:15472746
Long, L. R., Antani, S., Deserno, T. M., & Thoma, G. R. (2009). Content-Based
Image Retrieval in Medicine: Retrospective Assessment, State of the Art, and Future
Directions. International Journal of Healthcare Information Systems and Informatics,
4(1), 1–16. doi:10.4018/jhisi.2009010101 PMID:20523757
Manjunath, K. N., Renuka, A., & Niranjan, U. C. (2007). Linear models of cumulative
distribution function for content-based medical image retrieval. Journal of Medical
Systems, 31(6), 433–443. doi:10.100710916-007-9075-y PMID:18041275
Muller, Michoux, Bandon, & Geissbuhler. (2007). A review of content-based image
retrieval systems in medical applications—clinical benefits and future directions.
Intl. Journal of Medical Informatics, 73(1), 1–23.
Nanni, L., Brahnam, S., & Lumini, A. (2012). A simple method for improving local
binary patterns by considering non-uniform patterns. Pattern Recognition, 45(10),
3844–3852. doi:10.1016/j.patcog.2012.04.007
Ojala, T., Pietikainen, M., & Harwood, D. (1996). A comparative study of texture
measures with classification based on feature distributions. Pattern Recognition,
29(1), 51–59. doi:10.1016/0031-3203(95)00067-4
Papakostas, G. A., Koulouriotis, D. E., Karakasis, E. G., & Tourassis, V. D.
(2013). Moment- based local binary patterns: A novel descriptor for invariant
pattern recognition applications. Neurocomputing, 99, 358–371. doi:10.1016/j.
neucom.2012.06.031
Pereira, Fonseca-Pinto, Paiva, Tavora, Assuncao, & Faria (2020). Accurate
segmentation of desmoscopic images based on local binary pattern clustering.
International Convention on Information and Communication Technology, Electronics
and Microelectronics, Opatija, Croatia.
Seng & Mirisaee. (2009). Evaluation of a content-based retrieval system for blood
cell images with automated methods. Journal of Medical Systems, 35, 571–578.
Sorensen, Shaker, & Bruijne. (2010). Quantitative analysis of pulmonary emphysema
using local binary patterns. IEEE Trans. Med. Imaging, 29(2), 559-569.

135
A Content-Based Approach to Medical Image Retrieval

Tan, X., & Triggs, B. (2010). Enhanced local texture feature sets for face recognition
under difficult lighting conditions. IEEE Transactions on Image Processing, 19(6),
1635–1650. doi:10.1109/TIP.2010.2042645 PMID:20172829
Xue, Long, & Antani, Jeronimo, & Thoma. (2008). A Web-accessible content-
based cervicographic image retrieval system. Proceedings of the Society for Photo-
Instrumentation Engineers, 6919.
Xueming, Hua, Chen, & Liangjun. (2011). PLBP: An effective local binary patterns
texture descriptor with pyramid representation. Pattern Recognition, 44, 2502–2515.
Yao, C.-H., & Chen, S.-Y. (2003). Retrieval of translated, rotated and scaled color
textures. Pattern Recognition, 36(4), 913–929. doi:10.1016/S0031-3203(02)00124-3
Yurdakul, Subathra, & Georgec. (2020). Detection of Parkinson’s Disease from
gait using Neighborhood Representation Local Binary Patterns. Biomedical Signal
Processing and Control, 62.
Zakeri, F. S., Behnam, H., & Ahmadinejad, N. (2010). Classification of benign and
malignant breast masses based on shape and texture features in sonography images.
Journal of Medical Systems, 36(3), 1621–1627. doi:10.100710916-010-9624-7
PMID:21082222
Zhang, B., Gao, Y., Zhao, S., & Liu, J. (2010). Local derivative pattern versus local
binary pattern: Face recognition with higher-order local pattern descriptor. IEEE
Transactions on Image Processing, 19(2), 533–544.
Zhang, D., Yang, G., & Li, F. (2020). Detecting seam carved images using
uniform local binary patterns. Multimedia Tools and Applications, 79, 8415–8430.
doi:10.100711042-018-6470

136
137

Chapter 8
Correlation and Analysis
of Overlapping Leukocytes
in Blood Cell Images Using
Intracellular Markers and
Colocalization Operation
Balanagireddy G.
Rajiv Gandhi University of Knowledge Technologies, India & Dr. A. P. J. Abdul
Kalam Technical University, Ongole, India

Ananthajothi K.
https://orcid.org/0000-0002-6390-2082
Misrimal Navajee Munoth Jain Engineering College, India

Ganesh Babu T. R.
Muthayammal Engineering College, India

Sudha V.
Sona College of Technology, India

ABSTRACT
This chapter contributes to the study of uncertainty of signal dimensions within a
microscopic image of blood sample. Appropriate colocalization indicator classifies
the leukocytes in the region of interest having ragged boundaries. Signal transduction
has been interpreted using correlation function determined fluorescence intensity in
proposed work using just another colocalization plugin (JaCoP). Dependence between

DOI: 10.4018/978-1-7998-3092-4.ch008

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

the channels in the colocalization region is being analysed in a linear fashion using
Pearson correlation coefficient. Manders split, which gives intensity, is represented
in a channel by co-localizing pixels. Overlap coefficients are also being analysed
to analyse coefficient of each channel. Li’s intensity correlation coefficient is being
used in specific cases to interpret the impact of staining.

1. INTRODUCTION

Blood sample image usually contains the following: erythrocytes (RBC), leukocytes
(WBC) and platelets. The major classification on “White blood cells” are denoted
as follows, “Neutrophils”, “Eosinophils”, “Monocytes” and “Lymphocytes”. Each
of these subtype cells contributes to the usefulness in body defence. Hence each
subtype of cells are taken and classified as imaging dimensions in accordance to its
shape. The main limitation behind the four subtypes of leukocytes is that, if they are
clumped together they may reduce the accuracy of classification. The purpose of this
research is to interpret the biological relevance between specific classes of leukocytes
using colocalization procedures. Spatial point characteristics are visually evaluated
to provide protrusion of cells associated with the region of interest. The chapter is
organised as follows section 2 deals with the previous research of processing blood
samples using image processing. Section 3, discusses algorithm for segmenting
the class of leukocytes with JaCoP. Section 4 discusses result of medical image
segmentation using colocalization method. Section 5 concludes the overall work.

2. LITERATURE SURVEY

“Immunohistochemical” slide image registration accuracy depends on the field of view

of few cells. Registration accuracy is achieved with biomarker colocalization using
an elastix framework based on dynamic resolution levels (Moles Lopez et al., 2015).
The classification of leukocytes based on their shape and lobes of nucleus is given
as follows. It can be mononuclear which includes “Monocytes” and “Lymphocytes”.
The other contains granules named as granulocytes, which includes Neutrophils
and Eosinophils. The extraction of WBC cells from blood samples is followed by
separation of cytoplasm and nucleus thereby further enhanced classification has
been done in (Putzu et al., 2014). Marker controlled watershed has been used for
segmentation based on cell nucleus. Subsequently, classification has been done to
separate WBC and RBC (Miao & Xiao, 2018). However, misclassification may
result in improper movement of WBC leading to cell adhesion.

138
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

A review of colocalization techniques has been discussed with “Manders co-

occurrence” (MOC) which considered pixel intensity (Manders et al., 1993). The
limitation of MOC is that, it can be affected by useless signal. Similarly, Pearsons
Correlation Coefficient (PCC) based on interdependency may result that depended
on threshold values (Aaron et al., 2018). Localization using colour transformation
methods and subsequent segmentation of cytoplasm via region growing, watershed
and finally classification is done using support vector machine has been done
using bone marrow images (Liu et al., 2019). Leukocyte nuclei segmentation using
channel splitting with blue and green channel can reduce the time incurred to analyse
the blood cell image (Wang & Cao, 2019). Both spatial and spectral features for
segmenting the nucleus and leucktocyte region was discussed. Spatial feature used
morphological operations, whereas spectral features used support vector machine
(Duan et al., 2019). Feature classifying benign and malignant cells incorporating
convolutional neural network and statistical property has been discussed with Salp
Swarm Algorithm (Sahlol et al., 2020). But the work included, most features which
are relevant and excludes noise. Arbitrary shape of biomedical image has been done
using colocalization in a cell of interest with dilated radius(CIRCOAST) (Corliss et
al., 2019). Molecular interaction of irregular shape has been discussed using Geo-
coPositioning system. It combines an object based method to intensity based method.
Object based method provides resilience against noise and content of fluorescence
signal is estimated using intensity based method (Lavancier et al., 2020). Biological
processing with “Digital Lensless holographic microscopy” states that ImageJ plugin
has been used for interoperability for calculating numerical simulation (Trujillo et
al., 2020). Hence this work used ImageJ with JaCoP plugins for analysis.
Semantic segmentation has been used initially after pre-processing the blood cell
images the pixel level features are extracted using deep convolutional encoder and
decoder. The accuracy in classifying the RBC, WBC and platelets are better using
Intersection of Union and Boundary Score (Shahzad et al., 2020).

2.2 Problem Definition

Fluorescence spectra are not separated well in image acquisition. Misalignment of

signal with imperfect representation of leukocytes cells with nucleus and cytoplasm
along with its overlap can result in various quantitative values or volumetric changes
leading to wrong diagnosis. Thus colocalization of individual images has to be done
manually, to interpret the intracellular markers.

139
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

3. PROPOSED SYSTEM

3.1 Assumptions

The assumption is made such that the background is excluded and the region of
interest in an overlapping area, marked in leukocytes is taken for colocalization.

3.2 Algorithm

Step 1. Convert the given image into its corresponding colour across the red, green
and blue channel.
Step 2. Depending on the hue values through visual inspection the corresponding
green or red channel is taken for processing the leukocyte region.
Step 3. Segment the leukocyte region of interest manually by adjusting the wand
tool using appropriate threshold values. Crop this region of interest and display
as separate image.
Step 4. Segment different level of nuclei and cytoplasm inside the leukocyte region
from the cropped image.
Step 5. Correlation coefficients for interpreting the region of interest in the images
are Pearson’s correlation coefficient, Manderco localization coefficient, Overlap
coefficient. In addition,

Li’s intensity correlation coefficient is used in certain images.

Step 6. Coste’s mask (Costes et al., 2004) is being used with channel intensity as
in equation 1 and equation 2.

I 1 = C + ROI 1 (1)

I 2 = α + C + ROI 2 (2)

Notations used are denoted as follows, I1and I2denote the channel intensity of the
image 1 and image 2 respectively. Colocalized component is denoted by C, random
component in image 1 is ROI1and random component in image 2 is ROI2 and α is
“stoichiometry coefficient”.

140
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

4. RESULTS AND DISCUSSION

The Blood cell image has been obtained from (Mooney, 2018). Correlation functions
of images for colocalization has been analysed using “overlap coefficient”, “Mander
coefficient”, “Pearson coefficient” and “Li’s intensity correlation coefficient”(Bolte
& Cordelières, 2006). The acquired images are in JPEG format but are converted
into dtif for colocalization procedures. The darkest purple colour represents the
nucleus surrounded by light purple denoting the cytoplasm. Erythrocytes are being
represented in pink colour.

Table 1. Description of leukocytes cells used in this work.

Type of WBC Cells Number of Images Image Size

Width 640
Figure 1a, Input image is taken
Neutrophils Height 480
from (Mooney, 2018).
Depth 3
Width 640
Figure 9a, Input image taken from
Eosinophils Height 480
(Mooney, 2018).
Depth 3
Width 640
Figure 13a, Input image taken
Monocytes Height 480
from (Mooney, 2018).
Depth 3
Width 640
Figure 16a, Input image taken
Lymphocytes Height 480
from (Mooney, 2018).
Depth 3

A high amount of saturationis present in green channel in the image used for
analysis in the first image work. Figure 2 (a) shows the whole image of a leukocyte
region and the (b), (c) and (d) represents nucleus area.
The whole region of the cell is being correlated with its sub-region nucleus. There
is no colocalization that exhibits between the cells. The threshold valueis taken for
first ROI and second ROI is being set to116.
The figure 3 (a), shows the whole cell where nucleus is in graycolor and cytoplasm
in red color. Thefigure 3 (b), sub region of interest where graycoloris nucleus and
cytoplasm in red color with yellow boundaries.
The whole region of the cell is being correlated with its sub-region of nucleus.
There is no colocalization that exhibits between these cells. The threshold value
taken for First ROI is 121 and second ROI is set as 116.
The figure 4 shows Coste’s automatic threshold using whole area threshold as
160 and threshold (c) as 123.

141
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 1. (a) Neutrophil identified as purple colourimage in RGB format. (b)

Neutrophil image with green channel.(c) Neutrophil image with red channel.(b)
Neutrophil image with blue channel.

Figure 2. (a) First Region of Interest (ROI) withwhole cell within Neutrophil image
ingreen channel. (b) Second Region of Interest (ROI) Neutrophil image with green
channel.(c)Third Region of Interest (ROI) Neutrophil image with green channel. (d)
Forth Region of Interest (ROI) Neutrophil image with green channel.

142
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Table 2. Denotes the correlation coefficient used for figure 2.

Pearson coefficient 0.549

Overlap coefficient 0.551
Mander coefficient
0.318
(Image 2 (b) overlapping with 2 (a) fraction
Mander coefficient
0.226
(Image 2 (a) overlapping with 2(b) fraction using threshold values

Figure 3.

Table 3. Denotes the correlation coefficient used for figure 2(a) and 2 (c).

Pearson coefficient 0.368

Overlap coefficient 0.37
Mander coefficient
0.143
(orginal)
Mander coefficient
0.051
(Image 2 (a) overlapping with 2(c) fraction using threshold value)
Mander coefficient
0.292
(Image 2 (c) overlapping with 2(a) fraction using threshold value)

Figure 4. Coste’s mask for ROI-3 over whole cell area.

143
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Table 4. Denotes the correlation coefficient used for figure 2 (a) and 2(d).

Pearson coefficient 0.318

Overlap coefficient 0.32
Mander coefficient
0.108
(original)
Mander coefficient
0.057
(Image 2 (a) overlapping with 2(d) fraction using threshold value)
Mander coefficient
0.481
(Image 2 (d) overlapping with 2(a) fraction using threshold value)

The figure 5, shows Coste’s automatic threshold using whole area threshold as
160 and threshold as 125.
The whole region of the cell is being correlated with its sub-region of nucleus.
There is one colocalization that exhibits between these cells. The threshold value
taken for First ROI is 118 and second ROI is set as 104.
The figure 7 shows Costes automatic threshold using whole area threshold as
143 for image 6(a) and 120 threshold for image 6 (c).
The whole region of the cell is being correlated with its sub-region of nucleus.
There is one colocalization that exhibits between these cells. The threshold value
taken for First ROI is 118 and second ROI is set as 104.
The figure 8 shows Costes automatic threshold using whole area threshold as
143 for image 6(a) and 128 threshold for image 6 (b).
The whole region of the cell is being correlated with its sub-region of nucleus.
The whole region of the cell is being correlated with its sub-region of nucleus.
There is minimal colocalization that exhibits between these cells. The threshold
value taken for First ROI is94 and second ROI is set as 85.
The figure 11 shows Costes automatic threshold using whole area threshold as
94 for image 10(a) and 85 threshold for image 10(b).
The whole region of the cell is being correlated with its sub-region of nucleus.
There is minimal colocalization that exhibits between these cells. The threshold
value taken for First ROI is116 and second ROI is set as 97.
The figure 12 shows Costes automatic threshold using whole area threshold as
116 for image 10(a) and 97 threshold for image 10(c).
The figure 15 shows Costes automatic threshold using whole area threshold as
111 for image 14(a) and 102 threshold for image 14(b).
The whole region of the cell is being correlated with its sub-region of nucleus.
There is minimal colocalization that exhibits between these cells. The threshold
value taken for First ROI is111 and second ROI is set as 102.

144
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 5. Costes mask for ROI-4 over whole cell area.

Figure 6. (a) First Region of Interest (ROI) withwhole cell within Neutrophil image
in green channel. (b) Second Region of Interest (ROI) Neutrophil image ingreen
channel.(c) Third Region of Interest (ROI) Neutrophil image ingreen channel.

The whole region of the cell is being correlated with its sub-region of nucleus.
There is one colocalization that exhibits between these cells. The threshold value
taken for First ROI is 98 and second ROI is set as 72.The correlation coefficient
values used for figure 17(a) and 17(b) is denoted in table 10.

145
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Table 5. Denotes the correlation coefficient used for figure 6 (a) and 6(c).

Pearson coefficient 0.61

Overlap coefficient 0.614
Mander coefficient
0.397
(orginal)
Mander coefficient
0.063
(Image 6 (a) overlapping with 6(c) fraction using threshold value)
Mander coefficient
0.081
(Image 6 (c) overlapping with 6(a) fraction using threshold value)

Figure 7. Costes mask for image 6(a) with ROI-(a) and ROI (c) over whole cell area.

Table 6. Denotes the correlation coefficient used for figure 6 (a) and 6(b).

Pearson coefficient 0.548

Overlap coefficient 0.551
Mander coefficient
0.307
(original)
Mander coefficient
0.305
(Image 6 (a) overlapping with 6(b) fraction using threshold value)
Mander coefficient
0.91
(Image 6 (b) overlapping with 6(a) fraction using threshold value)

146
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 8. Costes mask for image 6(a) with ROI-(a) and ROI (b) over whole cell area.

Figure 9. (a) Eosinophils identified in purple color representation in RGB image.

(b) Eosinophils image with green channel.(c) Eosinophils image with red channel.
(d) Eosinophils image with blue channel.

147
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 10. (a) First Region of Interest (ROI) withwhole cell within Eosinophil image
withgreen channel. (b) Second Region of Interest (ROI) Eosinophil image with green
channel.(c) Third Region of Interest (ROI) Eosinophil image with green channel.

Table 7. Denotes the correlation coefficient used for figure 10(a) and 10(b).

Pearson coefficient 0.548

Overlap coefficient 0.551
Mander coefficient
0.307
(original)
Li’s intensity correlation coefficient 0.487

Figure 11. a, Costes mask for image 10 with ROI-(a)over whole cell area and ROI
(b) with nucleus.

148
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Table 8. Denotes the correlation coefficient used for figure 10(a) and 10(c).

Pearson coefficient 0.671

Overlap coefficient 0.675
Mander coefficient
0.49
(original)
Li’s intensity correlation coefficient 0.491

Figure 12. a, Costes mask for image 10 with ROI-(a)over whole cell area and ROI
(c) with nucleus.

Figure 13. (a) Monocytesimagein RGB image. (b) Monocytes image with green
channel.(c) Monocytes image with red channel.(d) Monocytes image with blue channel.

149
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 14. (a) First Region of Interest (ROI) withwhole cell of Monocytesimage in
green channel. (b) Second Region of Interest (ROI) sub cell of Monocytesimage in
green channel.

Figure 15. a, Costes mask for image 14 with ROI-(a)over whole cell area and ROI
(b) with nucleus.

Table 9. Denotes the correlation coefficient used for figure 14(a) and 14(b).

Pearson coefficient 0.507

Overlap coefficient 0.52
Mander coefficient
0.293
(original)
Li’s intensity correlation coefficient 0.460

150
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 16. (a) Lymphocyteidentified as whole image in RGB format. (b)

Lymphocyteimage with green channel.(c) Lymphocyteimage with red channel.(d)
Lymphocyteimage with blue channel.

Figure 17. (a) First Region of Interest (ROI) Lymphocyteidentified as whole imagein
green channel. (b) Second Region of Interest (ROI) sub cell of Lymphocyteimage
in green channel.

Table 10. Denotes the correlation coefficient used for figure 17(a) and 17(b).

Pearson coefficient 0.625

Overlap coefficient 0.629
Mander coefficient
0.45
(original)
Li’s intensity correlation coefficient 0.491

151
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Figure 18. a, Costes mask for image 17 with ROI-(a) over whole cell area and ROI
(b) with nucleus.

5. CONCLUSION

This research contributes to the colocalization of overlapping leukocytes and its

calibration with correlation function. Thus individual leukocytes are analysed
with co-occurrence .and clustered leukocytes and are visually inspected and their
impact of correlation is being analysed. Larger cells of leukocytes and ragged
boundaries which navigate through complex tissues are being examined visually
via correlating coefficients. The limitation behind this approach taken depends on
the brightness of computer screen which cannot be taken as reliable marker for
medical diagnosis. Future work will deal considering the impact of bio-molecules
using colocalizationtechniques using over restoration of nucleus in WBC cells.

REFERENCES

Aaron, J. S., Taylor, A. B., & Chew, T. L. (2018). Image co-localization–co-occurrence

versus correlation. Journal of Cell Science, 131(3), jcs211847. doi:10.1242/jcs.211847
PMID:29439158
Bolte, S., & Cordelières, F. P. (2006). A guided tour into subcellular colocalization
analysis in light microscopy. Journal of Microscopy, 224(3), 213–232. doi:10.1111/
j.1365-2818.2006.01706.x PMID:17210054
Corliss, B. A., Ray, H. C., Patrie, J. T., Mansour, J., Kesting, S., Park, J. H., & Peirce,
S. M. (2019). CIRCOAST: A statistical hypothesis test for cellular colocalization with
network structures. Bioinformatics (Oxford, England), 35(3), 506–514. doi:10.1093/
bioinformatics/bty638 PMID:30032263

152
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Costes, S. V., Daelemans, D., Cho, E. H., Dobbin, Z., Pavlakis, G., & Lockett, S.
(2004). Automatic and quantitative measurement of protein-protein colocalization in
live cells. Biophysical Journal, 86(6), 3993–4003. doi:10.1529/biophysj.103.038422
PMID:15189895
Duan, Y., Wang, J., Hu, M., Zhou, M., Li, Q., Sun, L., & Wang, Y. (2019).
Leukocyte classification based on spatial and spectral features of microscopic
hyperspectral images. Optics & Laser Technology, 112, 530–538. doi:10.1016/j.
optlastec.2018.11.057
Lavancier, F., Pécot, T., Zengzhen, L., & Kervrann, C. (2020). Testing independence
between two random sets for the analysis of colocalization in bioimaging. Biometrics,
76(1), 36–46. doi:10.1111/biom.13115 PMID:31271216
Liu, H., Cao, H., & Song, E. (2019). Bone marrow cells detection: A technique for the
microscopic image analysis. Journal of Medical Systems, 43(4), 82. doi:10.100710916-
019-1185-9 PMID:30798374
Manders, E. M. M., Verbeek, F. J., & Aten, J. A. (1993). Measurement of co-
localization of objects in dual colour confocal images. Journal of Microscopy,
169(3), 375–382. doi:10.1111/j.1365-2818.1993.tb03313.x
Miao, H., & Xiao, C. (2018). Simultaneous segmentation of leukocyte and erythrocyte
in microscopic images using a marker-controlled watershed algorithm. Computational
and Mathematical Methods in Medicine, 2018, 2018. doi:10.1155/2018/7235795
PMID:29681997
Moles Lopez, X., Barbot, P., Van Eycke, Y. R., Verset, L., Trépant, A. L., Larbanoix,
L., & Decaestecker, C. (2015). Registration of whole immunohistochemical slide
images: An efficient way to characterize biomarker colocalization. Journal of the
American Medical Informatics Association: JAMIA, 22(1), 86–99. doi:10.1136/
amiajnl-2014-002710 PMID:25125687
Mooney. (2018 April). Blood cell images Version 6. Retrieved May 23 2020 from
https://www.kaggle.com/paultimothymooney/blood-cells
Putzu, L., Caocci, G., & Di Ruberto, C. (2014). Leucocyte classification for leukaemia
detection using image processing techniques. Artificial Intelligence in Medicine,
62(3), 179–191. doi:10.1016/j.artmed.2014.09.002 PMID:25241903
Sahlol, A. T., Kollmannsberger, P., & Ewees, A. A. (2020). Efficient classification
of white blood cell leukemia with improved Swarm optimization of deep features.
Scientific Reports, 10(1), 1–11. doi:10.103841598-020-59215-9 PMID:32054876

153
Correlation and Analysis of Overlapping Leukocytes in Blood Cell Images

Shahzad, M., Umar, A. I., Khan, M. A., Shirazi, S. H., Khan, Z., & Yousaf, W.
(2020). Robust Method for Semantic Segmentation of Whole-Slide Blood Cell
Microscopic Images. Computational and Mathematical Methods in Medicine, 2020,
2020. doi:10.1155/2020/4015323 PMID:32411282
Trujillo, C., Piedrahita-Quintero, P., & Garcia-Sucerquia, J. (2020). Digital lensless
holographic microscopy: Numerical simulation and reconstruction with ImageJ.
Applied Optics, 59(19), 5788–5795. doi:10.1364/AO.395672 PMID:32609706
Wang, Y., & Cao, Y. (2019). Quick leukocyte nucleus segmentation in leukocyte
counting. Computational and Mathematical Methods in Medicine, 2019, 2019.
doi:10.1155/2019/3072498 PMID:31308855

154
155

Chapter 9
Enchodroma Tumor
Detection From MRI Images
Using SVM Classifier
G. Durgadevi
New Prince Shri Bhavani College of Engineering and Technology, India

K. Sujatha
Dr. M. G. R. Educational and Research Institute, India

K.S. Thivya
Dr. M.G.R. Educational and Research Institute, India

S. Elakkiya
Dr. M.G.R. Educational and Research Institute, India

M. Anand
Dr. M.G.R. Educational and Research Institute, India

S. Shobana
New Prince Shri Bhavani College of Engineering and Technology, India

ABSTRACT
Magnetic resonance imaging is a standard modality used in medicine for bone
diagnosis and treatment. It offers the advantage to be a non-invasive technique that
enables the analysis of bone tissues. The early detection of tumor in the bone leads on
saving the patients’ life through proper care. The accurate detection of tumor in the
MRI scans are very easy to perform. Furthermore, the tumor detection in an image
is useful not only for medical experts, but also for other purposes like segmentation
and 3D reconstruction. The manual delineation and visual inspection will be limited
to avoid time consumption by medical doctors. The bone tumor tissue detection
allows localizing a mass of abnormal cells in a slice of magnetic resonance (MR).
DOI: 10.4018/978-1-7998-3092-4.ch009

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

INTRODUCTION

Medical image processing is an important field of research as its outcomes are used
for the betterment of health issues. A tumor is an abnormal growth of tissues in any
part of the body. As the tumor grows, the abnormal tissue displaces healthy tissue.
There is a large class of bone tumor types which have different characteristics. There
are two types of bone tumors, Noncancerous (Benign) and Cancerous (Malignant).
The benign tumor grows very large and press on nearby tissues, once removed by
surgery, they don’t usually reoccur. Malignant tumor has a larger nucleus that looks
different from a normal cell’s nucleus and can also reoccur after they are removed.
Hence care as to be taken in order to completely avoid tumors. There are different
image modalities like X-ray, MRI, CT, PET SCANS has shown in figure 1.1. The
MRI imaging technique is the best because it has a higher resolution. Magnetic
resonance imaging (MRI) is a non-invasive medical system used to show 2D images
of the body. This technique is based on a process that uses highly charged magnetic
fields and radio waves to make images of the inside the body. It is an unharmed
method of obtaining images of the human body. Its data are most relevant and it helps
in early detection of tumors and precise estimation of tumor boundaries. Magnetic
resonance (MR) sequences such as T1-weighted, T2-weighted, contrast-enhanced
T1W and T2W, STIR (Short T1 inversion recovery), PD-Weighted series provide
different information. Thus MRI scans have a best non-invasive medical systems
used to show 2D images of the body. This technique is based on a process that used
highly charged magnetic fields to make images of the body. Hence MRI has more
than one methodology to classify images. These are atlas methods, shape methods,
fuzzy methods, and variations methods. New technology MRI are T1 weighted, T2
weighted and proton density weighted images.
The rest of the paper includes section 2 gives the brief glimpse of the relevant
work that was carried out all in the various fields of research. Section 3explains
segmentation process -thresholding and morphological operations. Section 4 includes
the proposed method with results and experimental results. Section 5 includes the
conclusions followed by future enhancements.

REVIEW OF LITERATURE

Sinan Onal et al. (Onal et al., 2014) proposed a method of automatic localization of
multiple pelvic bone structure on MRI, and they have used an SVM classification and
nonlinear regression model with global and local information, and they are presented
to automatically localize multiple pelvic bone Durgadevi et al. (Durgadevi & Shekhar,
2015) proposed a method of Identification of tumour using k-means algorithm.

156
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

The identification of breast cancer from the MRI images is made automatic using
K-means clustering and wavelet transform. The human perception at many times
may lead to erroneous diagnosis. Variation in diagnosis may produce adverse effect
on the patients. Hence to improve the accuracy this system is made automatic using
machine vision algorithms. Alan Jose et al. (Emran et al., 2015) described Brain
Tumour Segmentation Using K-Means Clustering and Fuzzy C-Means Algorithms
and Its Area Calculation; they have given Simple Algorithm for detection of range
and shape of tumour in brain MR Images. Normally the anatomy of the Brain can
be viewed by the MRI. MRI scanned image is used for the entire process. The MRI
scan is more comfortable than any other scans for diagnosis Deepak et al. (Jose
et al., 2014) discussed Comparative Study of Tumour Detection Techniques with
their Suitability for Brain MRI Images The canny edge detection technique defines
edges of the MRI image by using many parameter like thresholding, thinning etc.
canny with morphological operation like dilation, erosion etc., where simply applied
on it for getting better results, and fuzzy c-means method gives best results for
segmentation of Brain tumour in MRI images. M. Koch, et al,(Bagahel & Kiran,
2015) described automatically segmenting the wrist bones in the arthritis patients
using the k-means clustering process.

Figure 1. MRI SCAN

157
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

PROPOSED SYSTEM

Pre-processing is a basic step, which aims to improve the quality of an image by

removing the noise artefacts and increasing contrast. some pre-processing techniques
are image acquisition, histogram equalization filtering. Filtering is the basic step
among all. Filtering reduces the noise, saves the edges and smooth improper images
produced by the MRI. Figures 1.2 shows images with enchondroma.
We chose the bilateral filter. The bilateral filter is a filter for sharping and for
using nonlinear images. Since we use SVM classifier along with the bi lateral used to
detect the nonlinear images the bi lateral filter is also used to reduce the noise over the
images and also gives the contrast to the image. Figure 3 shows the bilateral output.

Figure 2. Enchondroma images

158
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

The input image is taken as a noisy one and then pre-processing is carried out
in previous step then the boundaries have to be calculated with the help of the
thresholding and morphological operation. Segmentation is the process of dividing
the images into regions. We are using two segmentation process thresholding and
morphological operations. Thresholding is the most common process, which divides
the images into a binary images.it is very productivity with images that have high
resolution. Figure 4 and 5 shows the results of thresholding and morphological
operations. Thresholding alone is not enough because in some cases it may case
false segments. That is false images. So, dilation and erosion is used. Which means
shrink and expand. Both the operations are helpful to detect the tumor area.

Figure 3. preprocessing technique

In Proposed, SVM Classifier used to segment the cancer detected portion. To

segment the portion, first have to filter out the acquired image based upon the
masking methodology. The Morphological function including dilation and erosion
method will be applied extracted throughout the filtered image. By the method of
morphological bounding box will be drawn over the affected portion. Then, the region
enclosed by bounding box will be splitted out separately with the SVM Classifier.
The block diagram in figure 6 explains the various blocks to execute the
segmentation process. In thatInput Bone tumor image is given as input in this module.
That is taking the affected tumor part from the MRI images and processing In this
stage we used two method of segmentation, they are morphological operation and
thresholding segmentation. Thresholding alone is not enough because in most cases

159
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

the images have artefacts or false segments.so erosion and dilation operations are
carried out to expand and shrink the region of interest

EXPERIMENTAL RESULTS

The following screenshot explains the classification done by SVM classifier and
the tumor region is separated fig 1.7shows the SVM classifier is used to classify
the segmented region, where as figure 1.8 shows the classified tumor detection.

Figure 4. Thresholding segmentation

CONCLUSION

The detection of Bone Tumour from the MRI images is taken away, and the images
that do not have a tumour or an unrelated images requires two main steps, namely
pre-processing and segmentation. In the pre-processing step, bilateral filters smooth
the images and remove the noise. We combine two segmentation along with it namely
morphological as well thresholding process. With various process. SVM classifier
is used to detect the tumor portions and finally segmentation is carried out. With

160
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

the SVM classifier the tumor regions are separated from the non-tumor sections.
The detection of enchondroma tumor from the MRI images using SVM classifier
is carried out along with the pre-processing techniques and segmentations process.
We develop an application to assess the performance of the proposed method via
MATLAB 2015R and in advanced.

Figure 5. Morphological operation

Figure 6. Proposed architecture

161
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

Figure 7. Classification

Figure 8. Segmented Tumor

162
Enchodroma Tumor Detection From MRI Images Using SVM Classifier

REFERENCES

Bagahel, D., & Kiran, K.G. (2015). Comparative study of tumor detection techniques
with their suitability for brain MRI images. Inlt. Jrl., 127(13).
Durgadevi & Shekhar. (2015). Identification of tumor using K-means algorithm.
Intl. Jrl. Adv. Res. Inn. Id. Edu, 1, 227-231.
Emran, Abtin, & David. (2015). Automatic segmentation of wrist bones in CT using
a statistical wrist shape pose model. Inlt. Jrl.
Jose, Ravi, & Sampath. (2014). Brain tumor segmentation a performance analysis
using K-means, fuzzy c-means and region growing algorithm. Inlt. Jrl., 2(3).
Onal, Susana, Paul, & Alferedo. (2014). Automated localization of multiple pelvic
bone structure in MRI. Intl Jrl.

163
164

Chapter 10
An Approach to Cloud
Computing for Medical
Image Analysis
M. P. Chitra
Panimalar Institute of Technology, India

R. S. Ponmagal
SRM Institute of Science and Technology, India

N. P. G. Bhavani
Meenakshi College of Engineering, India

V. Srividhya
Meenakshi College of Engineering, India

ABSTRACT
Cloud computing has become popular among users in organizations and companies.
Security and efficiency are the two major problems facing cloud service providers and
their customers. Cloud data allocation facilities that allow groups of users to work
together to access the shared data are the most standard and effective working styles
in the enterprises. So, in spite of having advantages of scalability and flexibility, cloud
storage service comes with confidential and security concerns. A direct method to
defend the user data is to encrypt the data stored at the cloud. In this research work,
a secure cloud model (SCM) that contains user authentication and data scheduling
approach is scheduled. An innovative digital signature with chaotic secure hashing
(DS-CS) is used for user authentication, followed by an enhanced work scheduling
based on improved genetic algorithm to reduce the execution cost.

DOI: 10.4018/978-1-7998-3092-4.ch010

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
An Approach to Cloud Computing for Medical Image Analysis

INTRODUCTION TO CLOUD COMPUTING

Cloud computing is a computing model, where a substantial pool of computing

frameworks are associated in the private or open systems, to give progressively
versatile foundation to the execution of Personal Computer (PC) application and
information stockpiling. With the approach of this innovation, the expense of the
computational procedure, application facilitating, content stockpiling and conveyance
is diminished altogether. The virtual pictures of the physical machines in the data
centers are provided to the clients. Virtualization is one of the principle ideas of
the cloud computing framework as it basically assembles the reflection over the
physical framework. Cloud computing is a prototype for enabling appropriate and
on-demand network access to a shared pool of computing resources that can be
rapidly stipulated and released with the cloud service provider interaction or minimal
management effort. The service provider provides different types of services such
as Software as a Service (SaaS), Platform as a Service (PaaS) or Infrastructure as
a Service (IaaS) to the customers across the world through the Internet. The cloud
resizes the virtualized hardware automatically. Cloud computing system provides
the technologies and tools to compute intensive parallel applications with affordable
prices when compared to the existing parallel computing techniques. The architecture
of cloud computing is shown in the Figure 1. The benefits of cloud computing
system are depicted in Figure 2. The advantages of the cloud computing system are
described below Ankita Yadav et.al., (2016).

• Cost effective
• On-demand services
• Remote access
• High efficiency and scalability
• Improved flexibility and reliability
• Maximum resilience without redundancy

TYPES OF CLOUD DEPLOYMENT AND SERVICE MODELS

Cloud Deployment Models

Cloud deployment models can be divided into four types: public, private, community
and hybrid cloud. Different types of deployment models are described below

• Public Cloud
• Private Cloud

165
An Approach to Cloud Computing for Medical Image Analysis

• Community Cloud
• Hybrid Cloud

Public Cloud

The public cloud allows the systems and services to be easily accessible to the
general users. Public cloud might be less secure because of its receptiveness, e.g.
email. Public cloud is a completely virtualized condition. Furthermore, suppliers
have a multi-inhabitant engineering that empowers clients or occupants to share
the figuring assets, for example, PCs, virtual machines, servers, applications and
capacity Yarlagadda et.al.,(2011). Information of each tenant is stored in the public
cloud. Notwithstanding, it stays segregated from different inhabitants. Public cloud
likewise depends on the high-transmission capacity arrange network to quickly
transmit information Masdari et.al., 92017). Public cloud stockpiling is commonly
excess, utilizing various data centers and cautious replication of document renditions.
Figure 3 demonstrates the public cloud.

Figure 1. Cloud Computing Architecture

166
An Approach to Cloud Computing for Medical Image Analysis

Figure 2. Benefits of cloud computing system

The main benefits of a public cloud service are:

• It prevents the need for the business organizations to invest and maintain their
own on-premises resources.
• High scalability to meet workload and user demands.
• Fewer wasted resources as the customers should only pay for the resources
they use.

Figure 3. Public cloud

167
An Approach to Cloud Computing for Medical Image Analysis

Private Cloud

The private cloud allows the computing systems and services to be accessible
within an organization or business enterprise. It offers expanded security due to its
private nature. Private cloud is a sort of cloud computing that conveys comparable
points of interest to the general population cloud, including adaptability and self-
benefit, yet through an exclusive engineering. In contrast to open clouds, which
convey administrations to numerous associations, a private cloud is committed
to the necessities and objectives of a solitary association Mehmood et.al., (2013).
Private clouds are sent inside firewalls and offer strong Information Technology
(IT) security for the association. On the off chance that a server farm foundation is
as of now accessible with the association the private cloud can be executed in-house.
For having in-house private clouds the association needs to put vigorously in running
and keeping up the framework which can result in critical capital consumption.
Figure 4 shows the private cloud Liao et.al.,(2013).

Figure 4. Private cloud

Advantages

• Better data control for the users and cloud service providers.
• As the cloud belongs to a single client, there will be high levels of security.
• As they are deployed inside the firewall of the organization’s intranet, it
ensures high efficiency and good network performance.
• Easy automation of the hardware and other resources by the company.

168
An Approach to Cloud Computing for Medical Image Analysis

Community Cloud

The community cloud allows the systems and services to be accessible by group
of organizations. A community cloud is a cloud benefit show that gives a cloud
registering solution for a predetermined number of people or associations that is
represented, overseen and anchored regularly by all the taking an interest associations
or an outsider overseen specialist organization Bace et.al., (2011). Community clouds
are a half and half type of private clouds assembled and worked particularly for
a focused on gathering. These people group have comparable cloud prerequisites
and their definitive objective is to cooperate to accomplish their business targets.
Community clouds are intended for organizations and associations chipping away at
joint tasks, applications, or research, which requires a focal cloud office for building,
overseeing and executing such undertakings, paying little mind to the arrangement
leased Bashir et.al.,(2014).

Figure 5. Community cloud

Advantages

• Highly secure private multi-tenant cloud computing

• Improved scalability
• Low cost by eliminating the infrastructure and software licenses needs
• Faster implementation, minimum development time and easy access to the
test environments

169
An Approach to Cloud Computing for Medical Image Analysis

• Better application, hosting, storage and backup

• Access to the distributed information and advanced analytics solutions

Hybrid Cloud

The Hybrid Cloud is a combination of the public and private cloud. Nonetheless,
the basic exercises are performed utilizing the private cloud, while the non-basic
exercises are performed utilizing the public cloud. A hybrid cloud is a model in which
a private cloud associates with public cloud framework, enabling an association
to arrange workloads across the two environments. A hybrid cloud organization
requires an abnormal state of similarity between the hidden programming and
administrations utilized by both the public and private clouds. Figure 6 depicts the
hybrid cloud Bharati et.al.,(2017), Werlinger et.al.,(2008) .

Advantages

• High scalability
• Improved Security
• Enhanced organizational operability
• Greater data accessibility
• Low cost requirement

Figure 6. Hybrid cloud

Service Models

Figure 7 depicts the cloud service models. Different types of service models are
described below

170
An Approach to Cloud Computing for Medical Image Analysis

• Infrastructure as a Service (IaaS)

◦◦ Software IaaS (For example: Google Compute Engine, IBM Smart
Cloud, AWS)
◦◦ Hardware IaaS (For example: Rackspace Cloud, AWS, Google Compute
Engine)
• Platform as a Service (PaaS)
• Software as a Service (SaaS)
• Data as a Service (DaaS)

Figure 7. Cloud Service Models

IaaS

Figure8 illustrates the IaaS model. IaaS provides access to fundamental resources
such as physical machines, virtual machines, virtual storage, etc. In an IaaS model,
a cloud specialist organization has the foundation parts generally present in on-
premises server farm, including servers, stockpiling and networking equipment,
and the virtualization or hypervisor layer. IaaS supplier likewise supplies a scope
of administrations to go with those framework parts. IaaS clients get to assets and
administrations through a Wide Area Network (WAN) for example, the web, and can
utilize the cloud supplier’s administrations to introduce the rest of the components of

171
An Approach to Cloud Computing for Medical Image Analysis

an application stack. Associations pick IaaS since usually less demanding, quicker
and more cost-productive to work an outstanding task at the hand without purchasing,
oversee and bolster the hidden foundation. With IaaS, a business can basically lease
or rent that framework from another business Mell et.al., (2011), Modi et.al.,(2013).

PaaS

PaaS is a cloud computing model in which a third-party provider delivers hardware

and software tools usually those needed for application development to the users over
the internet. PaaS provides the runtime environment for applications, development &
deployment tools, etc. A PaaS provider builds and supplies a resilient and optimized
environment on which users can install applications and data sets. Users can focus
on creating and running applications rather than constructing and maintaining the
underlying infrastructure and services Modi et.al., (2012), Oktay et.al., (2013). Instead
of obtaining the hardware downright, users pay for the IaaS on demand. Infrastructure
is scalable depending on processing and storage needs. Figure 9 shows the PaaS.

Figure 8. IaaS

172
An Approach to Cloud Computing for Medical Image Analysis

Figure 9. PaaS

Advantages of PaaS
PaaS offers additional features such as middleware, development tools and other
business tools offer more advantages like cut coding time, add development
capabilities without adding staff, develop for multiple platforms including mobile,
use sophisticated tools affordably, support geographically distributed development
teams and efficiently manage the application lifecycle Mehmood et.al., (2013),
Dhage et.al.,(2011).

SaaS

SaaS model allows usage of software applications by the end users. Figure 10 shows
the SaaS model. SaaS provides a complete software solution on a pay-as-per-go basis
from a cloud service provider. SaaS is a cloud computing offering that provides users
with access to the cloud-based software of vendor. Users do not install applications
on their local devices. Instead, the applications reside on a remote cloud network
accessed through the web or an Application Programming Interface (API). Through
the application, users can store and analyze data and collaborate on projects. SaaS
vendors provide users with software and applications via a subscription model.
SaaS providers manage, install or upgrade software without requiring any manual
intervention Shelke et.al., (2012), Modi et.al.,(2012).

173
An Approach to Cloud Computing for Medical Image Analysis

Figure 10. SaaS

DaaS

DaaS is an information provision and distribution model in which data files

including text, images, sounds, and videos are made available to customers over a
network, typically the Internet. DaaS offers convenient and cost-effective solutions
for customer- and client-oriented enterprises Patel et.al.,(2013), Lee et.al., (2011).
DaaS allows for the separation of data cost and usage from software or platform
cost and usage. DaaS is expected to facilitate new and more effective ways of
distributing and processing data. DaaS is closely related to Storage as a Service and
(SaaS) and may be integrated with one or both of these provision models. Figure
11 shows the layout diagram of DaaS Alharkan et.al., (2012), Xing et.al.,(2013).
The characteristics of cloud are

• On demand self-services
• Wide Network Access
• Resource Pooling
• Rapid Elasticity
• Dynamic Computing Infrastructure

174
An Approach to Cloud Computing for Medical Image Analysis

• IT Service-centric Approach
• Minimally or Self-managed Platform
• Consumption-based Billing
• Multi Tenancy
• Cost-effective
• Scalable

Figure 11. DaaS

SECURITY IN CLOUD COMPUTING

Security is a major concern in the cloud computing system. Companies such as

Amazon, Google and Microsoft are enhancing the services provided for their
users. The Cloud Service Providers (CSPs) are concerned about the non-adequate
security measures and aspects such as data integrity, control, audit, confidentiality
and availability. In the cloud computing environment, the consumers can access the
computational resources online at any time through Internet without managing the
original resources issues such as physical and technical management. Protection
of the private information of the user is a major concern in the cloud computing
environment. When deciding whether or not to move into the cloud, the cloud

175
An Approach to Cloud Computing for Medical Image Analysis

users would take into account factors such as service availability, security, system
performance, etc. The cloud users may lose physical control over their applications
and data. In cloud environment, network perimeters will no longer exist from the
perspective of cloud users, which renders traditional security protection mechanisms
such as firewalls not applicable to the cloud applications Modi et.al.,(2012) Kholidy
et.al.,(2012).
In the healthcare applications, cloud service providers and/or system administrators
may not be allowed to access sensitive data when providing improved data security
protection according to the corresponding compliances. It requires that cloud
service providers are able to provide necessary security services to meet security
requirements of the individual cloud user while abiding to the regulations and/or
compliances. Data protection is a crucial security issue for most of the organizations.
Before moving into the cloud, cloud users need to clearly identify data objects to be
protected and classify data based on their implication on security, and then define
the security policy for data protection as well as the policy enforcement mechanisms.
For most applications, data objects would include not only bulky data at rest in cloud
servers, but also data in transit between the cloud and the users can be transmitted
over the Internet or via mobile media. Data objects may also include user identity
information created by the user management model, service audit data produced
by the auditing model, service profile information used to describe the service
instances, temporary runtime data generated by the instances, and other application
data Chung et.al., (2013).
Different types of data would be of different value and hence have different
security implication to cloud users. User identity information can contain Personally
Identifiable Information (PII) and has impact on user privacy. Therefore, just
authorized users should be allowed to access user identity information. Service audit
data provide the evidences related to compliances and the fulfillment of Service
Level Agreement (SLA), and should not be maliciously manipulated. Service profile
information could help attackers locate and identify the service instances and should
be well protected. Temporary runtime data may contain critical data related to user
business and should be segregated during runtime and securely destroyed after
runtime Modi et.al., (2012).

Security Services

The basic security services for information security include assurance of data
Confidentiality, Integrity, and Availability (CIA). In the Cloud Computing system,
the issue of data security becomes more complicated because of the intrinsic cloud
characteristics. Potential cloud users were able to safely move their applications/
data to the cloud. A suit of security services are described as follows.

176
An Approach to Cloud Computing for Medical Image Analysis

Data Confidentiality Assurance

This service protects data from being disclosed to illegitimate parties. In Cloud
Computing, data confidentiality is a basic security service to be in place. Although
different applications may have different requirements in terms of what kind of data
need confidentiality protection, this security service could be applicable to all the
data objects Bace et.al., (2001).

Data Integrity Protection

This service protects data from malicious modification. When having outsourced
their data to remote cloud servers, cloud users must have a way to check whether
or not their data at rest or in transit are intact. Such a security service would be of
the core value to cloud users. When auditing cloud services, it is also critical to
guarantee that all the audit data are authentic since these data would be of legal
concerns Werlinger et.al.,(2008).

Guarantee of Data Availability

This service assures that data stored in the cloud are available on each user retrieval
request. This service is particularly important for data at rest in cloud servers and
related to the fulfillment of SLA. For long-term data storage services, data availability
assurance is of more importance because of the increasing possibility of data damage
or loss over the time Mell et.al., (2011).

Secure Data Access

This security service is to limit the disclosure of data content to authorized users. In
practical applications, disclosing application data to unauthorized users may threat
the cloud user’s business goal. In the critical applications, inappropriate disclosure
of sensitive data can have juristic concerns. For better protection on sensitive data,
cloud users may need fine-grained data access control in the sense that different
users may have access to different set of data.

Regulations and Compliances

In practical application scenarios, storage and access of sensitive data should obey
specific compliance. In addition to this, the geographic location of data would
frequently be of concern due to export-law violation issues. Cloud users should

177
An Approach to Cloud Computing for Medical Image Analysis

thoroughly review these regulation and compliance issues before moving their data
into the cloud.

Service Audition

This service provides a way for cloud users to monitor data access and is critical
for compliance enforcement. In the case of local storage, it is not hard to audit the
system. In Cloud Computing, it requires the CSP to support trustworthy transparency
of data access Yarlagadda et.al., (2011).

1ATTACKER MODEL

In the Cloud Computing system, the cloud users move applications from within their
enterprise/organization boundary into the open cloud. The cloud users lose physical
control over their data. In an open environment, the cloud users may confront all kinds
of attacks. Though there might be various categorization methods for the attacks,
it is useful to identify where these attackers come from and what kind of attacks
they can launch. Based on this criterion, the attackers are classified as insiders and
outsiders in the cloud computing environment Masdari et.al., (2017).

Insiders

The insiders refer to the subjects within the system. They could be malicious
employees with authorized access privileges inside the cloud user’s organization,
malicious employees at the side of CSP, and even the CSP itself. In practice, an
employee, at both the cloud user side and the CSP side, could become malicious
for reasons such as economic benefits. These insider attackers can launch serious
attacks such as learning other cloud users’ passwords or authentication information,
obtaining control of the virtual machines, logging all the communication of other
cloud users, and even abusing their access privilege to help unauthorized users gain
access to sensitive information. In practical deployments, the cloud users may have
to establish trust relationship with the CSPs.
The misbehavior of the cloud server can be anyone or the combination of the
following:

1. Potentially decide to hide data corruptions caused by server hacks or Byzantine

failures to maintain reputation.
2. Neglect to keep or deliberately delete some rarely accessed data files to save
resources.

178
An Approach to Cloud Computing for Medical Image Analysis

3. Acquire data information by eavesdropping and monitoring the network traffic.

4. Even collude with a small number of malicious users for the purpose of
harvesting the data file contents when it is highly beneficial.

Cloud users should thoroughly review all the potential vulnerabilities and protect
their assets on any intentional or in advertent security breaches. More specifically,
the users should be aware about the security services offered by the service providers
and implementation of these security services. Verification mechanisms should be
available to cloud users for verifying the security services provided by the CSPs. For
valuable and/or sensitive data, the cloud users may also have to implement their own
security protection mechanisms, e.g., strong cryptographic protection, in addition
to the security service offered by the cloud service providers Liao et.al., 2013).

Types of Insiders

Companies implement a sophisticated technology to monitor their employees but

it’s not always easy for them to distinguish between an insider and an outside attack.
Those who target and plan attacks from the outside might create strategies for
obtaining insider knowledge and access by either resorting to an existing employee,
or by making one of their own an insider.

Compromised Actors

Insiders with the access credentials or computing devices have been compromised
by an outside threat actor. These insiders are more challenging to address since the
real attack is coming from outside, posing a much lower risk of being identified.

Unintentional Actors

Insiders who expose data accidentally, such as an employee who accesses company
data through public Wireless Fidelity (Wi-Fi) without the unsafe knowledge are
unintentional actors. A large number of data breach incidents result from employee
negligence towards security measures, policies, and practices.

Emotional Attackers

Insiders who steal data or destroy company networks intentionally, such as a former
employee who injects malware or logic bomb in corporate computers on his last
day at work.

179
An Approach to Cloud Computing for Medical Image Analysis

Tech Savvy Actors

Insiders who react to challenges are tech savvy actors. They use their knowledge of
weaknesses and vulnerabilities to breach clearance and access sensitive information.
Tech savvy actors can pose some of the most dangerous insider threats, and are
likely to sell confidential information to external parties or black market bidders.

Outsiders

The percentage of external threats to an organization is very high. It includes well-

funded hackers, organized crime groups, and government entities. Attacks can be
either active or passive. An active attack generates packets or participates in the
network while a passive attack is dropping the network or tracking users. By moving
data into the cloud, users will lose their conventional network perimeters and expose
their data in an open system. Just like any other open systems, Cloud Computing
could be vulnerable to malicious attacks through the Internet. This is because Cloud
Computing usually does not limit the type of user when providing services. Malicious
attackers can easily log into the cloud and launch attacks. More specifically, outsider
attackers can launch both passive attacks such as eavesdropping the network traffic,
and active attacks like phishing legitimate users’ credential, manipulating network
traffic and probing the cloud structure. For some cloud services, outsider attackers
can launch very severe attacks by taking advantage of the system flaw. By launching
the cross virtual machine attacks, the attackers are able to monitor VMs from their
co-resident VMs and threaten their security. By subverting hypervisors, attackers
are even able to control the whole system stack above the hypervisor.
To address outsider attacks, cloud service providers have the responsibility to
secure their cloud infrastructure, isolate user application in the cloud, patch system
flaws timely, and notify the cloud users with any discovered security risks. Cloud
users should strictly abide to the security guidance when using cloud services for the
purpose of reducing the possibility of security breach. Cloud users need to negotiate
recovery and backup mechanism with service providers for better security protection.
Table 1 shows different types of attacks Masdari et.al.,(2017).

INTRUSION DETECTION IN CLOUD

Intrusion Detection System (IDS) is an essential component of defensive measures

protecting computer systems and network against harm abuse. It becomes a
crucial section in the Cloud computing environment. The main aim of IDS is to
detect computer attacks and provide the proper response. An IDS is defined as the

180
An Approach to Cloud Computing for Medical Image Analysis

technique that is used to detect and respond to intrusion activities from malicious
host or network. There are mainly two categories of IDSs, network based and host
based. In addition, the IDS can be defined as a defense system, which detects hostile
activities in a network. The key is to detect and possibly prevent activities that
may compromise system security, or some hacking attempt in progress including
reconnaissance/data collection phases that involve for example, port scans. One key
feature of the IDS is their ability to provide a view of unusual activity and to issue
alerts notifying administrators and/or blocking a suspected connection. Intrusion
detection is defined as the process of identifying and responding to malicious activity
targeted at computing and networking resources. In addition, IDS tools are capable
of distinguishing between insider attacks and external attacks.

Table 1. Attacks in Cloud Computing

Attacks Description
This includes multiple attacks, including brute force, common passwords and
dictionary attacks, which aim to obtain password of the user. The attacker can try to
Password guessing
guess a specific user’s password, try common passwords to all users or use an already
attack
made list of passwords to match against the password file, in their attempt to find a
valid password.
The attacker tracks the authentication packet and replays this information to get an
Replay attack
unauthorized access to the server.
The attacker passively puts himself between the user and the verifier during the
Man-in-the-middle
authentication process. Then, the attacker attempts to authenticate by pretending to be
attack
as the user to the verifier and the verifier to the user.
The attacker pretends to be the verifier to the user to obtain authentication keys or
Masquerade attack
data that may be used to authenticate fallaciously to the verifier.
Insider assisted The systems managers intentionally compromise the authentication system or thieve
attack authentication keys or relevant data of users.
It is a web based attack in which the attacker redirects the user to the fake website to
get passwords/ PIN of the user. Social engineering attacks that use fake emails, web
Phishing attack
pages and other electronic communications to encourage the user to disclose their
password and other susceptible information to the attacker.
The attacker spies the user’s movements to get his/her password. In this type of attack
Shoulder-surfing
the attacker observes the user; how he enters the password i.e. what keys of keyboard
attack
the user has pressed.
In DoS attack, an attacker overloads the target cloud system with service requests so
Denial of Service
that it stop responding to any new requests and hence made resources unavailable to
(DoS) attacks
its users.
An attacker tries to inject malicious service or virtual machine into the cloud. In this
Cloud Malware
type of attack, the attacker creates own malicious service implementation module
Injection Attack
(SaaS or PaaS) or virtual machine (IaaS), and add it to the Cloud system.
continues on following page

181
An Approach to Cloud Computing for Medical Image Analysis

Table 1. Continued

Attacks Description
An attacker attempts to compromise the cloud system by placing a malicious virtual
Side channel attack machine in close proximity to a target cloud server system and then launching a side
channel attack.
In this type of attack, all possible combinations of password apply to break the
Brute Force Attacks password. The brute force attack is generally applied to crack the encrypted
passwords that are saved in the form of encrypted text.
This attack tries to match the password with most occurring words or words of daily
Dictionary Attack
life usage.
The key loggers are the software programs which monitors the user activities by
Key Loggers
recording each and every key pressed by the user.
In wrapping attack, the attacker tries to insert the malicious element in the Simple
Object Access Protocol (SOAP) message structure in Transport Layer Service (TLS)
Wrapping attack
and copies the fake content of the message into the server to interrupt the server
operation.
In flooding attack, an adversary can easily create fake data and whenever the server is
Flooding attack
overloaded, it allocates the job to the nearest server.
The attacker steals the information of user account and password. In this attack,
Data stealing attack
confidential information user is lost by the activity of the challenger.
Eavesdropping
If an attacker can read the transmitted keys, an eavesdropping will happen.
attack
Spoofing attack An attacker makes an interrupt by changing routing information and keys.
When the server needs to retain the password for later authentication, the keys are
Privileged insider
probably being stolen by the adversary because the server can find out the new
attack
password.
Server An attacker masquerades as a legitimate user. To succeed the user impersonation
impersonation attack attack, an attacker has to generate a valid login message.
In this attack, an adversary which is plotted inside the member can modify the
Stolen-verifier attack
passwords or the verification tables stored in the server’s database.
Parallel session In this attack, an adversary applies messages in another authentication process to
attack replace the messages in the authentication operation.
Perfect forward This attack happens when an adversary is able to acquire the patient password or a
secrecy secret key, and it will still be able to compute previous session keys.
Resistance to server This type of attack can be completely solved by providing the mutual authentication
spoofing attack between user and server.
An adversary can reveal the identity through offline exhaustive guessing. The user’s
Identity guessing
identity is usually short and has a certain format. Hence, an adversary may find the
attack
identity (ID) within multinomial time by executing complete guessing.
When a challenge-response authentication system is used as the same protocol in
Reflection attacks both directions by each side to authenticate the other side, a reflection attack will
happen.

182
An Approach to Cloud Computing for Medical Image Analysis

Host-Based IDS (HIDS)

Host-based IDS involves software or agent components run on the server, router,
switch or network appliance. But, the agent versions must report to a console or can
be run together on the same host. Basically, HIDS provides poor real-time response
and cannot effectively defend against one-time catastrophic events. In fact, the HIDSs
are much better in detecting and responding to long term attacks such as data theft.
HIDS collect information from a particular host and analyze to detect intrusive
events. The information may be system logs or audit trails of operating system.
HIDS analyzes the information and if there is any change in the behavior of
system or program, it reports to the0 network manager that the system is under
attack. The effectiveness of HIDS can be improved by specifying the features that
provide it more information for detection. However, it requires more storage for
information to be analyzed. In the case of cloud computing network, it is possible
to deploy HIDS on hypervisor, VM or host to analyze the system logs, user login
information or access control policies and detect intrusion events. HIDS is capable
of analyzing encrypted traffic however; it is susceptible to DoS attack and can
even be disabled. HIDS are commonly used to protect the integrity of the software
Mehmood et.al., (2013).

Network-Based IDS (NIDS)

This type of IDS captures network traffic packets such as TCP, UDP and IPX/SPX
and analyzes the content against a set of rules or signatures to determine if a possible
event took place. False positives are common when an IDS system is not configured
or tuned to the environment traffic it is trying to analyze. Networks based IDS (NIDS)
capture the traffic of entire network and analyze it to detect possible intrusions
like port scanning, DoS attacks etc. NIDS usually performs intrusion detection by
processing the IP and transport layer headers of captured network packets. It utilizes
the anomaly based and signature based detection methods to identify intrusions.
NIDS collects the network packets and looks for their correlation with signatures
of known attacks or compares the user current behavior with their already known
profiles in real-time. Multiple hosts in the network can be secured from attackers
by utilizing a few properly deployed NIDSs. If run in the stealth mode, the location
of NIDS can be hidden from attacker. The NIDS is unable to perform analysis if
traffic is encrypted. In cloud environment, the attacks on hypervisor or VMs are
detected by positioning NIDS at the cloud server that interacts with external network.
However, it cannot detect attacks inside a virtual network contained by hypervisor.
Cloud provider is responsible for installing NIDS in the cloud Mehmood et.al.,(2013).
Figure 12 shows the host-based IDS and network based IDS.

183
An Approach to Cloud Computing for Medical Image Analysis

Figure 12. Host-based and Network-based IDS

Virtual Machine Monitor (VMM)/Hypervisor Based IDS

Hypervisor provides a platform for communication among VMs. Hypervisor based

IDSs is deployed at the hypervisor layer. It helps in analysis of available information
to detect the anomalous activities. The information is based on communication at
various levels like communication between VM and hypervisor, between VMs and
communication within the hypervisor based virtual network Mehmood et.al.,(2013).

Distributed IDS (DIDS)

DIDS comprises numerous IDSs that are deployed across a large network to monitor
the traffic for intrusive behavior. The participant IDSs can communicate with each
other or with a centralized server. Each of these individual IDSs has its own two
function components: detection component and correlation manager. Detection
component monitors the system or subnet and transmits the collected information
in a standard format to the correlation manager. Correlation manager combines
information from multiple IDS and generates high level alerts corresponding to an
attack. Analysis phase makes use of signature and anomaly based detection methods
hence DIDS can detect known and unknown attacks. In case of cloud, DIDS can
be located at any of two positions: at processing server or host machine Mehmood
et.al., (2013).

184
An Approach to Cloud Computing for Medical Image Analysis

DETECTION TECHNIQUES USED BY IDS

Signature Based IDS

Signature based detection is performed by comparing the information collected from

a network or system against a database of signatures. A signature is a predefined
set of rules or patterns that correspond to a known attack. This technique is also
known as misuse detection Liao et.ai.,(2013). It can efficiently detect known attacks
with negligible false alarms. Signature based method helps network managers with
average security expertise to identify intrusions accurately. It is a flexible approach
since new signatures can be added to database without modifying existing ones.
However, it is unable to detect unknown attacks Bace et.al., (2001). In the Cloud
environment, signature based intrusion detection method can be utilized at front-end
of the cloud for detection of the known attacks from external network. It can also
detect both internal and external intrusions if deployed at back end of the cloud.

Anomaly Based Detection

Anomaly based detection compares current user activities against preloaded profiles
of users or networks to detect abnormal behavior that may be intrusions. The profiles
may be dynamic or static and correspond to the expected or benign behavior of
users. To build a profile, regular activities of users, network connections, or hosts
are monitored for a specific period of time called as training period. Profiles are
developed using various features like failed login attempts, number of times a file
is accessed by a particular user over a particular time duration, CPU usage etc.
Anomaly based detection is effective against unknown attacks. An attack detected
by anomaly based technique can be used as a signature in signature based detection.
However it produces a large number of false alarms due to irregular network and
user behavior. Moreover, it also requires large data sets to train the system for the
normal user profiles.
Soft computing techniques used for intrusion detection are described below

1. Fuzzy Logic: It is based on probability, uses values ranging between 0 and 1,

and is used to define degree of anomaly in intrusion detection.
2. Artificial Neural Networks (ANN): In intrusion detection, ANN can be used
for generalization of data from imperfect data. It is also used to categorize
data as being normal or anomalous.
3. Support Vector Machines (SVM): SVM can be an effective way to detect
intrusive events in case of limited data samples, where data dimensions will
not change the accuracy.

185
An Approach to Cloud Computing for Medical Image Analysis

4. Association rules: This technique helps in creation of new signatures which

can be used to detect intrusions. Such intrusions consist of some known attacks
or variation of known attacks.
5. Genetic Algorithm (GA): The network features selected by using GAs can be
applied in other techniques which improves the detection accuracy of IDS.

Hybrid Detection

The efficiency of IDS can be significantly improved by combining signature

based and anomaly based techniques that are called as Hybrid detection technique.
The motivation behind this combination is the ability to detect both known and
unknown attacks using signature based and anomaly based detection techniques.
Hybrid IDSs can be divided into two categories: 1) sequence-based, in which either
anomaly detection or misuse detection is applied first, and the other one is applied
next; 2) parallel-based, in which multiple detectors are applied in parallel, and the
final decision is made based on multiple output sources. The most common type of
hybrid system is to combine misuse detection and anomaly detection together. In
such a hybrid system, the signature detection technique detects known attacks, and
the anomaly detection technique detects new or unknown attacks.

PROBLEM STATEMENT

IDS technology has not reached a level where it does not require human intervention.
Latest IDS technology offers some automation like notifying the administrator in
case of detection of a malicious activity, shunning the malicious connection for a
configurable period of time, dynamically modifying a router’s access control list in
order to stop a malicious connection etc. But it is still very important to monitor the
IDS logs regularly to stay on top of the occurrence of events. Monitoring the logs
on a daily basis is required to analyze the kind of malicious activities detected by
the IDS over a period of time. Today’s IDS has not yet reached the level where it
can give historical analysis of the intrusions detected over a period of time. This is
still a manual activity. The IDS technology works on the attack signatures. Attack
signatures are attack patterns of previous attacks. The signature database needs to
be updated whenever a different kind of attack is detected and the fix for the same
is available. The frequency of signature update varies from vendor to vendor.
The successful growth of the artificial intelligence techniques has placed a great
challenge of incorporating this new field in IDS. Use of neural networks can also
be effective in IDS. Their capability to process huge data and derive meaning and
patterns from it can be applied to find attacks. Gradually, it keeps on learning keeping

186
An Approach to Cloud Computing for Medical Image Analysis

track of previous penetrations and analyzing data for newer ones. As cloud sellers
utilize virtual machine innovation Host and system interruption attacks on remote
hypervisors are a genuine security challenge. DOS and Distributed (DDOS) attacks
are hurled to refuse assistance accessibility to complete clients. Protection of data
from the outsider inspector is another stress of cloud security. Cloud reviewing might
be a hard assignment to look at consistence of all the security approaches by the
vendor Bharati et.al.,(2017). An IDS is complex and provide many challenges for
security practitioners. IDS research has focused largely on improving the accuracy
of these systems and on providing support to practitioners during the ongoing task
of monitoring alerts and analyzing potential security incidents. The installation
and the initial configuration of the IDS can be challenging that they can serve as a
barrier to use Werlinger et.al.,(2008).

RESEARCH CONTRIBUTION

Detection of the intrusions and attacks through unauthorized users is one of the
major challenges for both cloud service providers and cloud users. The first phase
of the research work proposes a new IDS based on the combination of One-Class
Support Vector Machine (OC-SVM) network, and Artificial Bee Colony (ABC) to
detect anomalies in complex dataset. The hybrid OC-SVM algorithm is substandard
because of it is not able to effect the representation based learning in the middle
hidden layer. This approach was implemented for different datasets such as NSL-
KDD, KDD-CUP datasets. The experimental results showed improved accuracy in
intrusion detecting attacks by the unauthorized access.
Cloud data allocation facility allows a group of user to work together to access
and the shared data is one of the most standard and effective working styles in the
enterprises. Despite of the scalability and flexibility benefits, the cloud storage
service comes with the data confidentiality and the security concerns. A direct
method to defend the user data is to encrypt the data stored at the cloud. In this
research work, a Secured Cloud Model (SCM) that contains user authentication, and
data scheduling is suggested. An innovative Digital Signature with Chaotic Secure
Hashing (DS-CSH) is applied for user authentication, followed by an enhanced work
scheduling based on the improved genetic algorithm to reduce the execution cost. The
proposed SCM architecture yields better throughput, schedule success rate, lower
normalized schedule cost, end-to-end delay and packet loss rate. Thus, the proposed
SCM provides a secure environment with a higher QoS that can support more users.
In recent years, outlier detection is a well-investigated topic in data science Bashiret.
al.,(2014), Bharati et.al.,(2017). Learning rule to classify normal and anomalous
data without prior label is called unsupervised anomaly detection. One-class SVM

187
An Approach to Cloud Computing for Medical Image Analysis

(OC-SVM) is a most popular approach of unsupervised anomaly detection, in which

smooth boundary is constructed around the majority of probability mass of data
Werlinger et.al.,(2008). Nowadays, numerous feature extraction and feature selection
methods have been implemented by use with OC-SVM for high dimensional dataset
Ankita Yadav et.al.,(2016). Because of extraordinary success of deep auto encoder
networks used as a feature extractors in speech anomaly detection Yarlagadda et.al.,
(2011), several hybrid models uses combine method of OC-SVM with deep learning
for their applications Liao et.al., (2013). This hybrid model was proved that in terms
on accuracy and recognition rate for anomaly detection with maximum benefits using
two publically available CNN models ImageNet-MatConvNet-VGG-F (VGG-F) and
ImageNetMatConvNet-VGG-M (VGG-M) Bace et.al.,2001).

Robust Deep Autoencoders for Anomaly Detection

Along with the OC-SVM with deep learning hybrid approaches, a new novel method
for anomaly detection is to use deep autoencoders. Motivated by RPCA Bashir
et.al.,(2014) robust deep autoencoder method is introduced for anomaly detection
in unsupervised manner Bashir et.al., (2017). In which, the input data is decomposed
by two parts in Robust Deep Autoencoder (RDA) or Robust Deep Convolutional
Autoencoder (RCAE) as X = LD + S , where LD denotes the representation the
hidden layer of the autoencoder. The matrix S captures unwanted and outlier data
which are hard to estimate as shown in Equation 1. The decomposition is improved
by optimizing the objective function shown below:

θ,s
( )
min+ LD − Dθ E θ (LD ) + λS T 2,1
2
(2.1)

Such that X −LD − S = 0

The mentioned optimization issue is carried out by a hybrid method of back
propagation and Alternating Direct Method of Multipliers (ADMM) approach.

One Class SVM for Anomaly Detection

Apart from many algorithms, One-Class SVM (OC-SVM) is most commonly used
approach for outlier detection in an unsupervised manner Chung et.al.,(2013). This
approach is special case of SVM, which separate all the data by learning hyper-plane
from origin in a Reproducing Kernel Hilbert Space (RKHS) and maximize the space
between hyper-plane to the origin. Automatically in OC-SVM all the data points
are denoted as positive labeled data and the origin as the only negative labeled data.

188
An Approach to Cloud Computing for Medical Image Analysis

More exactly, consider the data set X without class information, and Φ (X )
denote as RKHS map function from the input space to the feature space F and a
hyper plane f (X n ) is constructed by

f (X n ) = wT Φ (X n ) − r (1)

This function is used to separate as many as mapped vectors Φ (X ) , n : 1, 2, …,

n
from the origin.
Where w − Norm perpendicular to the hyperplane and r − bias Factor of the
hyperplane.
By solving the below equation, the value of r and w can be obtained.

N
1 1 1
min w22 + ⋅
w ,r 2 v N ∑ max (0, r − w, Φ (X )) − r
n =1
n:
(2)

Where v ∈ (0, 1), is a parameter that control the hyper plane distance from origin
and number of points cross over the hyper-plane.
The proposed SCM-cloud framework is estimated with the novel algorithms
implemented for security. This section comprises two sub-sections: simulation
setup, and comparative analysis.

Simulation Setup

The proposed novel SCM-cloud architecture is implemented in the Java simulation

environment. The proposed SCM architecture is compared with OMC-RP, SecSDN-
Cloud, Opensec, AuthFlow, RPL-based SDN, FlowDefender, IC-PCP, BHEFT and
CWTS-GA. Table 2 defines the most important parameters used in the proposed
algorithm simulations. The number of users, Cloud Service Provider (CSP) and
difference time are varied to analyze the network performance. However, the
parameters are not limited to this list.

Performance Evaluation

The QoS efficiency is defined in terms of three important throughput parameters

as well as end-to-end delay and packet loss. Network throughput is defined as the
amount of data transmitted from user to cloud in a specified time period. The
throughput Tp , is calculated using the following formula:

189
An Approach to Cloud Computing for Medical Image Analysis

Table 2. Simulation Parameters

Number of users 10
Number of cloud providers 3
Queue type Drop Tail
Buffer capacity 3
Data Rate 100 Mbps
Transmission Interval 2 seconds
Simulation time 30 seconds

NPs
Tp = (3)
T

Here, T represents the time interval of transmission, NPs denotes the number
of packet to be transferred. Figure 13 shows that the proposed SCM model show
the gradual improvement of the throughput when simulation time is increased. The
throughput of the SCM model is maximum of about 1.186%, 60.47% and 80.24%
than the SecSDN-Cloud, Opensec and AuthFlow approaches, respectively.

Figure 13. Performance Comparison of Throughput

190
An Approach to Cloud Computing for Medical Image Analysis

CONCLUSION

The overview of cloud computing for medical image analysis gives a brief outline
of the existing intrusion detection approaches for the cloud computing environment
in the field of medical image processing. The new IDS based on the combination
of OC-SVM network, and ABC to detect anomalies in complex dataset. A new
intrusion detection system (IDS) is proposed based on a combination of a one-class
Support Vector Machine (OC-SVM) network, and artificial bee colony (ABC) to
detect anomalies in complex dataset.The experimental results showed improved
accuracy in intrusion detecting attacks by unauthorized access. A new Secure Cloud
Model (SCM) framework for medical image analysis using digital signature with
chaotic secure hashing and Work scheduling based on improved Genetic algorithm.
An innovative digital signature with chaotic secure hashing (DS-CS) is used for
user authentication, followed by an enhanced work scheduling based on improved
genetic algorithm to reduce the execution cost.

REFERENCES

Aggarwal, C. C. (2015). Outlier analysis. Data Mining, 237-263. doi:10.1007/978-

3-319-14142-8_8
Alharkan, T., & Martin, P. (2012). IDSaaS: Intrusion detection system as a service in
public clouds. Proceedings of the 2012 12th IEEE/ACM International Symposium on
Cluster, Cloud and Grid Computing (ccgrid 2012), 686-687. 10.1109/CCGrid.2012.81
Bace & Mell. (2001). NIST special publication on intrusion detection systems.
Booz-Allen and Hamilton Inc.
Bashir, U., & Chachoo, M. (2014). Intrusion detection and prevention
system: Challenges & opportunities. Computing for Sustainable Global
Development (INDIACom), 2014 International Conference on, 806-809. 10.1109/
IndiaCom.2014.6828073
Bharati, M., & Tamane, S. (2017). Intrusion detection systems (IDS) & future
challenges in cloud based environment. Intelligent Systems and Information
Management (ICISIM), 240-250. doi:10.1109/ICISIM.2017.8122180
Chung, C.-J., Khatkar, P., Xing, T., Lee, J., & Huang, D. (2013). NICE: Network
Intrusion Detection and Countermeasure Selection in Virtual Network Systems. IEEE
Transactions on Dependable and Secure Computing, 10(4), 198–211. doi:10.1109/
TDSC.2013.8

191
An Approach to Cloud Computing for Medical Image Analysis

Dhage, S. N., Meshram, B., Rawat, R., Padawe, S., Paingaokar, M., & Misra, A.
(2011). Intrusion detection system in cloud computing environment. Proceedings
of the International Conference & Workshop on Emerging Trends in Technology,
235-239. 10.1145/1980022.1980076
Kholidy, H. A., & Baiardi, F. (2012). CIDS: A framework for intrusion detection in
cloud systems. Information Technology: New Generations (ITNG), Ninth International
Conference on, 379-385. 10.1109/ITNG.2012.94
Lee, J.-H., Park, M.-W., Eom, J.-H., & Chung, T.-M. (2011). Multi-level intrusion
detection system and log management in cloud computing. Advanced Communication
Technology (ICACT), 13th International Conference on, 552-555.
Liao, H.-J., Lin, C.-H. R., Lin, Y.-C., & Tung, K.-Y. (2013). Intrusion detection
system: A comprehensive review. Journal of Network and Computer Applications,
36(1), 16–24. doi:10.1016/j.jnca.2012.09.004
Masdari, M., & Ahmadzadeh, S. (2017). A survey and taxonomy of the authentication
schemes in Telecare Medicine Information Systems. Journal of Network and Computer
Applications, 87, 1–19. doi:10.1016/j.jnca.2017.03.003
Mehmood, Y., Shibli, M. A., Habiba, U., & Masood, R. (2013). Intrusion detection
system in cloud computing: challenges and opportunities. 2013 2nd National
Conference on Information Assurance (NCIA), 59-66. 10.1109/NCIA.2013.6725325
Mehmood, Y., Shibli, M. A., Habiba, U., & Masood, R. (2013). Intrusion detection
system in cloud computing: challenges and opportunities. 2nd National Conference
on Information Assurance (NCIA), 59-66. 10.1109/NCIA.2013.6725325
Mell & Grance. (2011). The NIST definition of cloud computing. Academic Press.
Modi, C., Patel, D., Borisaniya, B., Patel, H., Patel, A., & Rajarajan, M. (2013). A
survey of intrusion detection techniques in cloud. Journal of Network and Computer
Applications, 36(1), 42–57. doi:10.1016/j.jnca.2012.05.003
Modi, C. N., Patel, D. R., Patel, A., & Muttukrishnan, R. (2012). Bayesian Classifier
and Snort based network intrusion detection system in cloud computing. In Computing
Communication & Networking Technologies (pp. 1–7). ICCCNT. doi:10.1109/
ICCCNT.2012.6396086
Modi, C. N., Patel, D. R., Patel, A., & Muttukrishnan, R. (2012). Bayesian Classifier
and Snort based network intrusion detection system in cloud computing. Third
International Conference on Computing Communication & Networking Technologies
(ICCCNT), 1-7.

192
An Approach to Cloud Computing for Medical Image Analysis

Modi, C. N., Patel, D. R., Patel, A., & Rajarajan, M. (2012). Integrating signature
apriori based network intrusion detection system (NIDS) in cloud computing.
Procedia Technology, 6, 905–912. doi:10.1016/j.protcy.2012.10.110
Nikolai, J., & Wang, Y. (2014). Hypervisor-based cloud intrusion detection system.
In Computing (pp. 989–993). Networking and Communications.
Oktay, U., & Sahingoz, O. K. (2013). Proxy network intrusion detection system
for cloud computing. International Conference on Technological Advances in
Electrical, Electronics and Computer Engineering (TAEECE), 98-104. 10.1109/
TAEECE.2013.6557203
Oktay, U., & Sahingoz, O. K. (2013). Proxy network intrusion detection system
for cloud computing. Technological Advances in Electrical, Electronics and
Computer Engineering (TAEECE) International Conference on, 98-104. 10.1109/
TAEECE.2013.6557203
Patel, A., Taghavi, M., Bakhtiyari, K., & Celestino Júnior, J. (2013). An intrusion
detection and prevention system in cloud computing: A systematic review. Journal of
Network and Computer Applications, 36(1), 25–41. doi:10.1016/j.jnca.2012.08.007
Shelke, M. P. K., Sontakke, M. S., & Gawande, A. (2012). Intrusion detection system
for cloud computing. International Journal of Scientific & Technology Research,
1, 67–71.
Werlinger, R., Hawkey, K., Muldner, K., Jaferian, P., & Beznosov, K. (2008).
The challenges of using an intrusion detection system: is it worth the effort?
Proceedings of the 4th symposium on Usable privacy and security, 107-118.
10.1145/1408664.1408679
Xing, T., Huang, D., Xu, L., Chung, C.-J., & Khatkar, P. (2013). Snortflow: A
openflow-based intrusion prevention system in cloud environment. Research and
Educational Experiment Workshop (GREE), 89-92. 10.1109/GREE.2013.25
Yadav, A., & Kumar, N. (2016). A Survey of Authentication Methods in Cloud
Computing. International Journal of Innovative Research in Computer and
Communication Engineering, 4, 19529–19533.
Yarlagadda, V. K., & Ramanujam, S. (2011). Data security in cloud computing.
Journal of Computer and Mathematical Sciences, 2, 1–169.

193
194

Chapter 11
Segmentation of Spine
Tumour Using K-Means and
Active Contour and Feature
Extraction Using GLCM
Malathi M.
Rajalakshmi Institute of Technology, India

Sujatha Kesavan
Dr. M. G. R. Educational Research Institute of Technology, India

Praveen K.
Chennai Institute of Technology, India

ABSTRACT
MRI imaging technique is used to detect spine tumours. After getting the spine image
through MRI scans calculation of area, size, and position of the spine tumour are
important to give treatment for the patient. The earlier the tumour portion of the
spine is detected using manual labeling. This is a challenging task for the radiologist,
and also it is a time-consuming process. Manual labeling of the tumour is a tiring,
tedious process for the radiologist. Accurate detection of tumour is important for
the doctor because by knowing the position and the stage of the tumour, the doctor
can decide the type of treatment for the patient. Next, important consideration in
the detection of a tumour is earlier diagnosis of a tumour; this will improve the
lifetime of the patient. Hence, a method which helps to segment the tumour region
automatically is proposed. Most of the research work uses clustering techniques
for segmentation. The research work used k-means clustering and active contour
segmentation to find the tumour portion.

DOI: 10.4018/978-1-7998-3092-4.ch011

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

INTRODUCTION

A spine tumour is abandoned growth of cells, which was found in the spinal cord.
It grows uncontrollably. It may cancerous (Malignant) or non-cancerous (Benign).
It may cause neurologic problems and in some cases it produces paralysis. It may
occur in the region of the spine. The two types of tumour are primary and secondary.
The primary tumour starts in the spinal cord and secondary spreads to another part
of the spine. Based on the region of the spine it may occur in cervical, thoracic
lumber and sacrum. Based on the location of the spine it may be classified into three
types like Intradural-extramedullary, intramedullary and extradural. (Yezzi et al.,
1997) The clear and accurate visual arrangements of an internal organ of our body
have been generated using various medical imaging modalities like CT (Computed
Tomography), MRI (Magnetic Resonance Imaging). This can be used to provide
the internal organizations of bone and skin. The various diseases of the human body
should be identified with the help of Medical imaging techniques. It can be used to
generate an actual structure of the human body to detect the abnormalities. MRI,
CT, Ultrasound, Positron Emission Tomography (PET), etc. were the different kinds
of medical imaging techniques.
(Hai, S, Fuyong Xing & Lin Yang 2016) Demonstrates the various brain imaging
techniques. For the CT images the tomography is the word, which originates from
two Greek words like tomos and graphia. The word tomos represents slice or section
and graphia represents the picture. From this, it will understand that CT provides
the detailed structure of internal organ of the human body. CT utilizes X-rays to
reproduce the internal organization of the human body. After CT imaging, the
reconstruction of an image depends on the X-ray absorption profile.
One of the dynamic and flexible radio imaging technique was MRI. The technique
uses electromagnetic radiation to acquire the internal structure of the human body.
The abnormalities in the soft region were found by the invasive MRI imaging
methods. The technology helps the physicians to find the abnormalities in chest,
lungs, bones etc. Unlike X–ray MRI does not uses harmful radiation. During MRI
imaging the human body aligns the hydrogen atoms of the body.
X-rays are electromagnetic waves which are used to provide useful information
about the human body. The X-ray absorption profile will differ for every tissue.
Dense tissue seems like white on CT film and soft tissue looks like gray. During
CT imaging techniques the appearance of the Lungs is black because the hollow
space within the lungs is filled with air. Unlike X-ray CT does not use the dangerous
radiation, it affects the human body. CT is one of the best medical imaging technique
and it helps to diagnose the diseases in various human body parts like Brain, Pelvis,
Liver, Chest, Abdomen, and Spine etc. Hence the suggested method utilizes MRI
Imaging Techniques to find tumour in the spinal cord.

195
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

RELATED STUDIES

Manual segmentation refers the human operator or physician performs segmentation

and labeling of an image by hand. The separation is performed in a slice by slice
method on a 3-D volumetric image. Depending on the artifacts present in the
medical image makes the segmentation is an easy or difficult process. But manual
segmentation requires a long time to complete the task. Recent days the CT and MRI
imaging technique is mostly used in the diagnosis, treatment planning and clinical
studies require computers to assist the radiologist experts. In order to perform the
segmentation of a large amount of images with the same accuracy, the computer
aided diagnosis was implemented
(Shan Shen et al., 2005) The author proposed technique in which noises are
present in MRI brain images due to poor operating performance, disturbances,
surroundings and poor equipment maintenance. This noise affects the accuracy of
the segmentation process. Many conventional algorithms are available for tumour
segmentation. Clustering is grouping of pixels depends on the intensity value of a
pixel. Conventional clustering method mainly based on the intensity value of a pixel,
which leads to poor segmentation. Hence segmentation of the tumour is improved
by using a neighborhood attraction method, which will provide comparative position
and structures of neighbouring pixels. The proposed methodology does not use
intensity as a single parameter to perform segmentation of tumour, but the intensity
of neighbouring pixels was also taken into account for segmentation of tumour.
(Li Hong Juang& Ming Ni Wu 2010) The author used conventional clustering
method for MRI brain tumour segmentation. The author describes the segmentation
process on a colour brain image. Most of the segmentation process is executed on
gray image. But the proposed work by the author is performed on RGB image. The
segmentation process on colour image provides a good accuracy and also reduces
the iteration time or time required.
(Madhukumar, S &Santhiyakumari, N 2015) The author used histogram guided
initialization to do qualitative comparison between Fuzzy C-means (FCM) and
k-Means segmentation. The accuracy of the above mentioned methods depends on its
ability to distinguish dissimilar tissue classes, independently. The research discloses
that FCM detects the vasogenic edema and the white matter as a only tissue class
and correspondingly gray matter and necrotic focus, besides this FCM is better to
designate these areas when compared with K means clustering.
(Georges et al., 1999) The author discussed and introduced a Deformable model in
order to obtain the boundary information of lesions. It may be calculated effectively
with the help of snake model in active contour segmentation. It helps the radiologist
to find the spatial relationship between every pixel.

196
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

(Andrew et al., 2020) The author uses deep learning technology to perform
segmentation from MRI images. The manual segmentation of affected region from
an input image is tedious process. Hence the research work uses the automatic
segmentation in order to improve the accuracy and overall efficiency of the process.
Recent years CNN is used for automatic segmentation because it has the ability to
handle large data set and provides better accuracy by comparing with the existing
segmentation methods. The CNN performance is evaluated using state of art results.
(Hyunseok Seo &Masoud Badiei Khuzani, 2020) The author discussed various
machine learning algorithms for segmentation of medical and natural images. The
author compared different machine learning algorithms like Markov random field,
K- means clustering, random forest etc., the conventional segmentation yields less
accurate then compared to deep learning techniques. But it is simple to implement
and have less complex structures. The research work also compares the various
learning architectures like ANN, CNN and RNN. Recent deep learning techniques
need a several hyper parameter tuning. Small changes in hyper planeparameters
yieldsmany changes in the network output

MATERIALS AND METHODS

Existing Method

MRI Brain-imaging technique is used to detect the brain tumour. After getting the
Brain image through MRI scans calculation of area, size and position of the brain
tumour are important to give the treatment for the patient. Earlier days the tumour
portion of the brain detected using manual labeling. This is a challenging task for the
radiologist and also it is a time- consuming process. Manual labeling of the tumour
is fatigue, tedious process for the radiologist. Accurate detection of brain tumour
is important for the doctor because by knowing the position and the stage of the
tumour, the doctor can decide the type of treatment for the patient. Next important
consideration in the detection of a tumour is earlier diagnosis of a tumour; this will
improve the lifetime of the patient. From this motive while analyzing the spine
tumour there is no good segmentation methods. Hence the article is initiated to
start with the conventional segmentation process like k means clustering followed
by active contour segmentation. But this method provides false segmentation in the
presence of noise and also it is difficult to assign the K – value. The proposed work
overcomes the drawback with suitable filters to remove nose and false segmentation
is avoided by active contour segmentation in order to partition the affected portion.
Further the image underdone for the feature extraction.

197
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Proposed Method

Thereare many types of segmentation methods used in medical images are thresholding
method, edge detection method based technology, region based technology, clustering
based technology. The proposed research work uses clustering based segmentation
due to its simplicity, accuracy and easy implementation. Recent years many types of
machine learning algorithms were developed like Artificial neural Networks (ANN’s),
Convolutional Neural networks (CNN’s), Recurrent neural networks(RNN’s) and
Deep neural networks (DNN’s). The future research work focused on developing the
segmentation using CNN and DNN. Since it has the ability to process large amount
of data. Before learning the recent machine learning algorithms it is necessary to
know the fundamental concepts on segmentation. Hence we proposed spine tumour
segmentation using conventional K means and active contour segmentation. In the
presence of noise K means algorithms provides false or over segmentation. The
traditional clustering based segmentation provides acceptable efficiency; it may be
improved by using CNN, DNN and ANN in future research work.
The segmentation of abnormal portions was calculated for many organs like
brain, breast, lungs etc., All segmentation process uses clustering methods. From
this motivation we applied automatic Spine tumour segmentation by implementing
K-means clustering and active contour segmentation process.
(Eman et al., 2015) finally the performance of the segmentation techniques is
measured by the number of GLCM features. The extracted features helps to classify
the abnormal portion is malignant or benign. The flow diagram is shown in figure 1.

Image Preprocessing

The After getting MRI Image preprocessing is the first step of segmentation process,
which helps to reduce noise and artifacts from the acquired image. The 0.02 amount
of salt and pepper noise is added to the acquired image. It can be replaced by median
filter. Since we have many types of filters to remove noise, but median filter was
mostly used due to the following reasons.

• The edge information should be preserved during preprocessing

• The median value is estimated by considering the value of an all pixels which
is available in the image.
• Nonlinear digital filtering method is used.

The following steps should be followed to calculate the median value. Consider
the particular pixel, for which the median pixel is calculated by organizing all the
pixels which is near to that. For an even number of pixels median value is calculated

198
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

by taking the average of two center pixel values is calculated to find median value.
Similarly for Odd value the center value of the pixels are median. Finally the
superiority of the MRI, spinal image is enhanced by conserving the edges.

Figure 1. Flow diagram of proposed model

K- Means Clustering

The flow diagram for K-means algorithm are denoted in the following figure
The algorithm is presented by Macqueen in the year 1997. It is an unsupervised
algorithm. It begins by randomly allocating the K total number of cluster center.
The cluster is calculated and it is named as centroid. Compare the every pixel with
neighboring pixels.
(Bjoern 2016 et al.,) Further, each pixel has shifted to nearest cluster center,
which has the shortest distance among all. This procedure is continuing till the center
converges. The several steps of the algorithms are described as follows.

199
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Step1: Arbitrarily choose the C cluster center.

Step 2: Evaluate Euclidean distance for all pixel to cluster centre.
Step 3: Each pixel is assigned to particular cluster, which has the smallest distance
Step 4: The algorithm helps to diminish the squared error,

Xi –Vi is the Euclidean distance between Xi, Vi.

C is the number of clusters.
Ci is the number of data points in the ithcluster. Next, calculate the cluster centre
by using the following formula

1 Ci
C=
Ci ∑ j =1
Xi

Figure 2. Flow chart for K means algorithm

200
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

The advantages are

• Implementation is easy.
• Easy to understand

The disadvantages are

• Care should be taken to choose the Proper K value.

• It is immune to noise and artifacts.

Clustering finds numerous applications in a variety of areas like image processing,

data mining, Image retrieval, pattern recognition, Image segmentation etc.

REGION BASED ACTIVE CONTOUR SEGEMENTATION

Active Contours

Active contouris one of the segmentation methods; it uses energy forces and constraints
for separation of the pixels of interest from the image for further processing. It
uses snake model, gradient vector flow snake model, balloon model and geometric
contours. Active contour model performs the segmentation on MRI, CT, PET and
SPECT images. The early diagnosis and detection of abnormalities of affected region
can be evaluated with the help of active contour models in 3-D imaging. This model
provides accurate results for 3-D CT and MRI images when comparing with other
methods. Segmentation of fine structures from the affected object in an image is
possible with the help of active contour models.
Region based segmentation check for the similarity of pixels based on the
properties like intensity, color and texture. These variation findings can be used
by active contour segmentation. For object detection, active contour model uses
33 types of curve evolution. Active contour is also named deformable model. It
is presented by Kass et al in 2-D space, and it can be upgraded for 3 –D space by
Terzopoulos et al. An Active contour or snakes used in 2D space, and balloons used
in 3 D space. In the presence of external force, the parametric curve transfers inside
the image to detect the boundaries of the object. (Georges et al., 1999) Discussed
the geometry; physics related information related of each pixels can be obtained by
deformable model. The above mentioned properties provide the variation of shape
over space and time.

201
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Step 1: Place the Active contour or snake neighboring to the abnormal region.
Step 2: In the presence of inside and outside forces produced in the image the snake
is relocated immediately to the target by an iterative process.
Step 3: Estimate the energy function of the forces.
Step 4: The main purpose of this technique is to reduce the energy function. The
data can be smoothened by internal forces and the contours are shifted to next
to the region of interest.

Region Properties

Eman et al. (2015) shows that the particular portion of an image has several properties
like perimeter, area, boundary box, major axis, centroid, eccentricity, convex, area,
etc. The property of the certain region represents the mathematical feature of the
specific portion of an image.
A region in an image can have many properties like area, perimeter, boundary box,
major axis length, minor axis length, centroid, eccentricity, filled image, orientation,
convex area, convex image, pixel list, solidity, Euler number, filled area, extrema,
subarray. Fundamentally the Region Properties indicate the mathematical features
of a particular region of image.

Area

It is the total amount of pixels in this specified portion. It is returned as a scalar

Perimeter

It is defined as the distance around boundary of the specific area, defined as a scalar.
The perimeter is estimated by measuring the distance among the all adjacent pair of
pixels around the boundary of the specific portion. In vase of discontinuous region
the command provides unexpected value.

FEATURE EXTRACTION USING GLCM

(Hai, S et al., 2016) The organization of pattern is made to be easy with the help
of feature extraction. It provides the information which is related to the shape of
an image. It reduces the number of resources required to classify the pattern. The
proposed work uses GLCM feature (Gray Level Co-occurrence Matrix) extraction.

202
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Figure 3. Results of clustering and active contour segmentation

GLCM matrix provides a statistical method which examines the texture of pixel by
considering a spatial relationship. (Mohammad Fathy et al., 2019) It characterizes the
quality of an image by estimating how frequently the pair of pixels with a particular
value occurs in the specified spatial relationship of an image.
A GLCM is a matrix, in which has equal number of rows and columns for gray
level of an image. The matrix element P (i, j | Δx, Δy), in which the two pixels has
relative frequency, separated by a distance (Δx, Δy) with the neighbourhood one
with intensity ‘j’.
For variations in the gray levels of ‘i’ and ‘j’ at a certain displacement distance
d and at a specific angle ө m the matrix element P (i, j | d, ө) provides the second
order statistical probability values.
GLCM is very sensitive to dimensions. The proposed extract some of the feature
using GLCM. It can be listed as follows

203
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Contrast

The difference in intensity contrast between the pixel and to adjacent pixels over
the entire image is represented by contrast.

C = i − j p (i, j )
2
(1)

Correlation

The relationship between the pixels and its neighbourhood pixel is represented by
correlation. The range of the value lies between (-1, 1). The correlation value is 1
for the positively correlated image, -1 for the negatively correlated image.

(i − µ )( j − µ )(p (i, j ))
i j
corr = ∑ (2)
σi σ j

µi, µj - Mean, σi , σ j - Standard deviation

Pi, Pj- Partial probability function

Energy

Energy is defined as a sum of squared elements in the GLCM (Gray Level Co-
occurrence Matrix) Energy is also mentioned as the quantity of uniformity. The
range of the value is between [0 1], for a constant, the energy value is 1

E = ∑P (i, j )
2
(3)
i, j

Entropy

It is used to represented as the uncertainty of the textural image.

L −1

h = −∑plk (log p )
(4)
2 lk
k =0

204
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

plk= probability of the Kthlevel

The extracted features can be used by different classifiers like in such a way to
help the neural network system to classify the abnormality of tumour affected portion.

Table 1. Feature extraction table

Input Images Contrast Correlation Energy Entropy

Image 1 0.205 0.742 0.8422 0.153
Image 2 0.901 0.756 0.924 0.531
Image 3 0.029 0.688 0.893 0.026
Image 4 0.291 0.7801 0.883 0.368

RESULTS AND DISCUSSION

Many of the brain tumour segmentation methods uses K means, Fuzzy C means
for separating tumour regions from the affected regions. While doing segmentation
using above two segmentation there is a possibility of over segmentation and false
segmentation happening in the presence of noise. Hence from the detailed review
hybrid clustering helps to overcome the disadvantages along with the BPN or SVM
classifier. It provides the efficiency of 93.28%. Further to perform the segmentation
on 3-D images we used active contour segmentation for our proposed work. The
accuracy and overall performance of the segmentation was improved with the help
of CNN. From this motivation the research work uses K means and Active contour
segmentation methods for spine tumour segmentation.

CONCLUSION

Initially the proposed work starts with a conventional segmentation algorithm

called K-means clustering. Since much of the research work for segmentation of
tumour portion is designed using clustering techniques. Because the method is
easy to understand and implement. But it finds many disadvantages in the presence
of artifacts, which occurs during MRI scan. From the motivation we have started
the work to find the tumor in the spinal cord is done by K means clustering. The
disadvantage of this technology is overcome through the active contour model. Since
it is compulsory to find the exact boundary of affected portions for feature extraction.

205
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

REFERENCES

Bjoern, H. M., & Koen, V. L. (2016). A Generative Probabilistic Model and

Discriminative Extensions for Brain Lesion Segmentation With Application to Tumor
and Stroke. IEEE Transactions on Medical Imaging, 35(4), 933–946. doi:10.1109/
TMI.2015.2502596 PMID:26599702
Eman, A. M., Mohammed, E., & Rashid, A. L. (2015). Brain tumor segmentation
based on a hybrid clustering technique. Egyptian Informatics Journal, 16(1), 71–81.
doi:10.1016/j.eij.2015.01.003
Fathy, M., Keshk, M., & El Sherif, A. (2019). Surgical management and outcome
of intramedullary spinal cord tumour. Egyptian Journal of Neurosurgery., 34(2),
2–7. doi:10.118641984-019-0028-9
Georges, B. A. (1999). Model Creation and Deformation for the Automatic
Segmentation of the Brain in MR Images. IEEE Transactions on Biomedical
Engineering, 46(11), 1346–1356. doi:10.1109/10.797995 PMID:10582420
Hai, S., Xing, F., & Yang, L. (2016). Robust Cell Detection of Histopathological
Brain Tumor Images Using Sparse Reconstruction and Adaptive Dictionary
Selection. IEEE Transactions on Medical Imaging, 35(6), 1575–1586. doi:10.1109/
TMI.2016.2520502 PMID:26812706
Hai, S., Xing, F., & Yang, L. (2016). Robust Cell Detection of Histopathological
Brain Tumor Images Using Sparse Reconstruction and Adaptive Dictionary
Selection. IEEE Transactions on Medical Imaging, 35(6), 1575–1586. doi:10.1109/
TMI.2016.2520502 PMID:26812706
Juang, L. H., & Ming, N. W. (2010). MRI brain lesion image detection based on
color-converted K-means. Science Direct Measurement, 43(7), 941–949.
Madhukumar, S., & Santhiyakumari, N. (2015). Evaluation of k-Means and fuzzy
C-means segmentation on MR images of brain. Egyptian Society of Radiology and
Nuclear Medicine, 46(2), 475–479. doi:10.1016/j.ejrnm.2015.02.008
Murathoti Varshini, Barjo, & Tigga. (2020). Spine Magnetic Resonance Image
Segmentation Using Deep Learning Techniques. 2020 6th International Conference
on Advanced Computing and Communication Systems (ICACCS).
Seo, H., & Khuzani, M. B. (2020). Machine learning techniques for biomedical
image segmentation: An overview of technical aspects and introduction to state-of-
art applications. American Association of Physicists in Medicine, 45(5), 148–167.
PMID:32418337

206
Segmentation of Spine Tumour Using K-Means and Active Contour and Feature Extraction

Shen, S., Sandham, W., Granat, M., & Sterr, A. (2005). MRI Fuzzy Segmentation
of Brain Tissue using Neighborhood Attraction with Neural-Network Optimization.
IEEE Transactions on Information Technology in Biomedicine, 9(3), 459–497.
doi:10.1109/TITB.2005.847500 PMID:16167700
Yezzi, A. J., Kichenassamy, S., Kumar, A., Olver, P., & Tannenbaum, A. (1997). A
geometric snake model for segmentation of medical imagery. IEEE Transactions on
Medical Imaging, 16(2), 199–209. doi:10.1109/42.563665 PMID:9101329

207
208

Chapter 12
A Survey on Early
Detection of Women’s
Breast Cancer Using IoT
P. Malathi
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

A. Kalaivani
Saveetha School of Engineering, India & Saveetha Institute of Medical and
Technical Sciences, Chennai, India

ABSTRACT
The internet of things is probably one of the most challenging and disruptive concepts
raised in recent years. Recent development in innovation and availability have
prompted the rise of internet of things (IoT). IoT technology is used in a wide scope
of certified application circumstances. Internet of things has witnessed the transition
in life for the last few years which provides a way to analyze both the real-time data
and past data by the emerging role. The current state-of-the-art method does not
effectively diagnose breast cancer in the early stages. Thus, the early detection of
breast cancer poses a great challenge for medical experts and researchers. This
chapter alleviates this by developing a novel software to detect breast cancer at a
much earlier stage than traditional methods or self-examination.

DOI: 10.4018/978-1-7998-3092-4.ch012

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
A Survey on Early Detection of Women’s Breast Cancer Using IoT

INTRODUCTION

Breast cancer the second leading cause of death for women. Breast cancer is cancer
that develops in breast cells. Typically, the cancer forms in either the lobules or
the ducts of the breast. Lobules are the glands that produce milk, and ducts are the
pathways that bring the milk from the glands to the nipple. Cancer can also occur in
the fatty tissue or the fibrous connective tissue within your breast. Among women,
breast cancer is the most second most common cancer diagnosed, after skin cancer,
and the second leading cause of cancer deaths, after lung cancer. On average, 1 in 8
women will develop breast cancer in their lifetimes. About two-thirds of women with
breast cancer are 55 or older. Most of the rest are between 35 and 54. Fortunately,
breast cancer is very treatable if you spot it early. Localized cancer (meaning it
hasn’t spread outside your breast) can usually be treated before it spreads. Once the
cancer begins to spread, treatment becomes more complicated. It can often control
the disease for years.

Symptoms of Breast Cancer

In Early stages, breast cancer may not cause any symptoms. In many cases, a tumor
may be too small to be felt, but an abnormality can still be seen on a mammogram.
If a tumor can be felt, the first sign is usually a new lump in the breast that was
not there before. However, not all lumps are cancer. Each type of breast cancer can
cause a variety of symptoms.

• A breast lump or tissue thickening that feels different than surrounding tissue
and has developed recently
• Breast pain
• Red, pitted skin over your entire breast
• Swelling in all or part of your breast
• A nipple discharge other than breast milk
• Bloody discharge from your nipple
• Peeling, scaling, or flaking of skin on your nipple or breast
• A sudden, unexplained change in the shape or size of your breast
• Inverted nipple
• Changes to the appearance of the skin on your breasts
• A lump or swelling under your arm

209
A Survey on Early Detection of Women’s Breast Cancer Using IoT

Treatment of Breast Cancer

Breast cancer is treated in several ways. It depends on the kind of breast cancer
and how far it has spread. People with breast cancer often get more than one kind
of treatment.

• Surgery: An operation where doctors cut out cancer tissue.

• Chemotherapy: Using special medicines to shrink or kill the cancer cells. The
drugs can be pills you take or medicines given in your veins, or sometimes
both.
• Hormonal therapy: Blocks cancer cells from getting the hormones they need
to grow.
• Biological therapy: Works with your body’s immune system to help it fight
cancer cells or to control side effects from other cancer treatments.
• Radiation therapy: Using high-energy rays (similar to X-rays) to kill the
cancer cells.

BREAST CANCER SCREENING TEST

Digital Mammogram

A mammogram is an X-ray of the breast. Mammograms are the best way to find
breast cancer early, when it is easier to treat and before it is big enough to feel or
cause symptoms. Having regular mammograms can lower the risk of dying from
breast cancer. At this time, a mammogram is the best way to find breast cancer for
most women.

Magnetic Resonance Imaging (MRI)

A breast MRI uses magnets and radio waves to take pictures of the breast. MRI is
used along with mammograms to screen women who are at high risk for getting
breast cancer. Because breast MRIs may appear abnormal even when there is no
cancer, they are not used for women at average risk.

Clinical Breast Exam

A clinical breast exam is an examination by a doctor or nurse, who uses his or her
hands to feel for lumps or other changes. The benefit of screening is finding cancer
early, when it’s easier to treat. Harms can include false positive test results, when

210
A Survey on Early Detection of Women’s Breast Cancer Using IoT

a doctor sees something that looks like cancer but is not. This can lead to more
tests, which can be expensive, invasive, time-consuming, and may cause anxiety.

Early Detection of Women Breast Cancer Techniques

Newer diagnostic techniques such as sestamibi scans, optical imaging and molecular
diagnostic techniques look promising, but need more investigation into their use.
Their roles will appear clearer in coming years, and they may prove to be of help
in further investigating lesions that are indeterminate on standard imaging. Other
upcoming techniques are contrast-enhanced mammography and tomosynthesis.
These may give additional information in indeterminate lesions, and when used in
screening they aid in reducing recall rates, as shown in recent studies. Tomography
has a role in detecting local disease recurrence and distant metastasis in breast
cancer patients. Computer-aided detection (CAD) is a software technology that has
become widespread in radiology practices, particularly in breast cancer screening
for improving detection rates at earlier stages. Many studies have investigated
the diagnostic accuracy of CAD. The current level of performance for the CAD
systems is encouraging but not enough to make CAD systems standalone detection
and diagnose clinical systems. Unless the performance of CAD systems enhanced
dramatically from its current level by enhancing the existing methods. Traditionally,
doctors have relied on mammograms to detect changes in breast tissue that could
indicate the growth of cancerous tissue. Unfortunately, mammograms are not always
accurate, particularly for women with dense breast tissue. However, an innovative
new “Internet of Things” (IoT). In the future, that home may be filled with IoT-
enabled devices that could transmit patient-generated health data to their doctors.
The information obtained from these devices could include vitals such as heart rate,
pulse ox, and respiratory rate.
In addition, IoT-enabled pillboxes, appliances, and even toothbrushes could also
generate a plethora of useful data. Yet other devices will detect time in bed, falls,
and even gait. All of this information will give clinicians (and family members) a
better idea of how patients are faring at home. For example, if IoT devices detect
that the patient hasn’t left their bed in a number of days nor opened their pill box
in a week, the system could alert their physician to take appropriate measures to
check on their patient.

Women Breast Cancer Detection Using Wearable Technology

In an innovative IoT wearable technology application, Cyrcadia Health is developing

a bra that could alert women to the early signs of breast cancer. A method to detect
breast cancer early. The accurate and efficient diagnosis of breast cancer is extremely

211
A Survey on Early Detection of Women’s Breast Cancer Using IoT

necessary for recovery and treatment in early stages in IoT healthcare environment.
However, that is not where they’re spending the majority of their day. That place
is their home. An iTBra and a vision of using wearable technology to drastically
improve for less money as well. In the future, that home may be filled with IoT-
enabled devices that could transmit patient-generated health data to their doctors.
The information obtained from these devices could include vitals such as heart rate,
pulse ox, and respiratory rate. In addition, IoT-enabled pillboxes, appliances, and
even toothbrushes could also generate a plethora of useful data. Yet other devices
will detect time in bed, falls, and even gait.
All of this information will give clinicians (and family members) a better idea of
how patients are faring at home. For example, if IoT devices detect that the patient
hasn’t left their bed in a number of days nor opened their pill box in a week, the system
could alert their physician to take appropriate measures to check on their patient.

Figure 1. Workflow for IoT bra

RELATED WORK

Internet of Things World Forum, Now-a-days hearing a lot about the transformational
value of the Internet of Things (IoT) across many industries manufacturing,
transportation, agriculture, smart cities, retail, and finance. So many new solutions
are on display that help organizations either save or make money. But in healthcare,
IoT can actually do more than that, it has the potential to save lives.
Lucia Arcarisi et. al.,(2019) in their research study, a non-invasive wearable device
designed to mimic the process of breast self-examination. It uses pressure sensing
textiles and thus increase the confidence and self-awareness of women. Combined
with other screening methods, the device can increase the odds of early detection
for better prognosis. The research work demonstrates that it can detect nodules in
much the same way as does the human hand during breast self-examination.

212
A Survey on Early Detection of Women’s Breast Cancer Using IoT

Muhammad Hammad Memon et. al., (2019) proposed machine learning-based

diagnostic system which effectively classifies the malignant and benign people
in the environment of IoT. The proposed system performance is excellent due to
the selection of recursive feature selection algorithm. The implementation of the
proposed system is very reliable in all aspects of IoT healthcare for breast cancer.
Md. Milon Islam et.al.,(2020) proposed a smart healthcare system in IoT
environment that can monitor a patient’s basic health signs and room condition of
the patients in real-time. In this system, five sensors are used to capture the data
from hospital environment named heart beat sensor, body temperature sensor, room
temperature sensor, CO sensor, and CO2 sensor. The condition of the patients is
conveyed via a portal to medical staff, where they can process and analyze the
current situation of the patients. The developed prototype is well suited for healthcare
monitoring that is proved by the effectiveness of the system.
Javier Andreu-Perez et. al., (2015) present the milestones and recent developments
of different generations of pervasive sensing applications for health monitoring. The
opportunities of pervasive health monitoring through data linkages with other health
informatics systems including the mining of health records, clinical trial databases,
multi-omics data integration and social media. Technical advances have supported
the evolution of the pervasive health paradigm towards preventative, predictive,
personalised and participatory medicine.
M. Sung et. al., (2005) describe LiveNet, a flexible wearable platform intended for
long-term Ambulatory health monitoring with real-time data streaming and context
classification. LiveNet is a stable, accessible system that combines inexpensive,
commodity hardware;, a flexible sensor interconnection bus and a powerful, light-
weight distributed sensing, classification, and inter-process communications software
architecture to facilitate the development of distributed real-time multi-modal and
context-aware applications. The paper demonstrate the power and functionality of
this platform by describing a number of health monitoring applications using the
LiveNet system in a variety of clinical studies.
Kim Gau Ng et. al.,(2012) proposed Cadi ThermoSENSOR skin-contact
thermometer measures body temperature continuously and transmits readings
wirelessly to a central server. This study evaluated the ThermoSENSOR against
ear temperatures (ETs) measured by a Braun ThermoScan ear thermometer and
axillary temperatures (ATs) measured by a Terumo digital clinical thermometer.
These results suggest that the TTs were comparable to the ETs and ATs.
This study employed a recognized protocol of activities both pre-operatively,
and at regular intervals up to twenty-four weeks post-total knee arthroplasty. The
results suggest that a wearable miniaturised ear-worn sensor is potentially useful in
monitoring post-operative recovery, and in identifying patients who fail to improve
as expected, thus facilitating early clinical review and intervention.

213
A Survey on Early Detection of Women’s Breast Cancer Using IoT

Direct interfacing of nanosensors onto biomaterials could impact health quality

monitoring and adaptive threat detection. The paper demonstrates the integration
onto a tooth for remote monitoring of respiration and bacteria detection in saliva.
Overall, this strategy of interfacing graphene nanosensors with biomaterials represents
a versatile approach for ubiquitous detection of biochemical targets.
M.-Z. Po et. al., (2010) evaluated the choice of electrode material by comparing
conductive fabric with Ag / AgCl electrodes and discuss the limitations found. The
evidence given in this work is a viable alternative to the traditional palmar sites
for EDA measurements. Our device offers the unprecedented ability to perform
comfortable, long-term, and in situ assessment of EDA. This paper opens up
opportunities for future investigations that were previously not feasible, and could
have far-reaching implications for diagnosis and understanding of psychological or
neurological conditions.
S.Patel (2009) presents the results of a pilot study to assess the feasibility of using
accelerometer data to estimate the severity of symptoms and motor complications in
patients with Parkinson’s disease. A support vector machine (SVM) classifier was
implemented to estimate the severity from accelerometer data features. SVM-based
estimates were compared with clinical scores derived via visual inspection of video
recordings taken while patients performed a series of standardized motor tasks. The
outcome of the proposed work is a thin, comfortable device technology that can
softly laminate on to the surface of the skin to enable advanced, multifunctional
operation for physiological monitoring in a wireless mode.
H. Zhou et. al., (2008) presents a new human motion tracking system using two
wearable inertial sensors that are placed near the wrist and elbow joints of the upper
limb. Each inertial sensor consists of a tri-axial accelerometer, a tri-axial gyroscope
and a tri-axial magnetometer. The turning rates of the gyroscope are utilised for
localizing the wrist and elbow joints on the assumption that the two upper limb
segment lengths. Experimental results demonstrate that this new system, compared
to an optical motion tracker, has RMS position errors that are normally less than
0.01 m, and RMS angle errors that are 2.5-4.8 degrees.
E. S. Sazonov (2011) presents a novel wearable sensor capable of very accurate
recognition of common postures and activities. The shoe sensor was tested in
nine adults performing sitting and standing postures and while walking, running,
stair ascent/descent and cycling. Support vector machines (SVMs) were used for
classification. A fourfold validation of a six-class subject-independent group model
showed 95.2% average accuracy of posture/activity classification on full sensor set
and over 98% on optimized sensor set. The proposed methods are implemented by
using signals from sensor-embedded shoes called smart shoes. Each smart shoe has
four GCF sensors installed between the cushion pad and the sole. The GCF sensor
applies an air pressure sensor connected to an air bladder. A gait monitoring system

214
A Survey on Early Detection of Women’s Breast Cancer Using IoT

that integrates the proposed methods is shown in this paper and verified for both
normal and abnormal gaits.
Flynn (2013) specified Rheumatoid Arthritis (RA) is a disease which attacks
the synovial tissue lubricating skeletal joints. This systemic condition affects the
musculoskeletal system, including bones, joints, muscles and tendons that contribute
to loss of function and Range of Motion (ROM). Traditional measurement of arthritis
requires labour intensive personal examination by medical staff which through their
objective measures may hinder the enactment and analysis of arthritis rehabilitation.

IOT Wearable for Breast

In an innovative IoT wearable technology application, Cyrcadia Health is developing

a bra that could alert women to the early signs of breast cancer. Cancer cells exhibit
abnormal temperature patterns, because the increased activity of the cancer cells
generates heat. The bra has built-in sensors that read cell temperatures and transmit
the data in real-time to a patient database. This Wi-Fi enabled garment contains
16 sensors that can detect changes in the wearer’s breasts. After wearing the bra
for two hours, the data is transmitted directly to the patient’s physician. This data
is then paired with a predictive algorithm that analyzes the information for known
risk factors. While it’s no replacement for an annual mammogram, this IoT bra
could alert patients and physicians to changes between their yearly appointments.
An abnormal reading triggers an alert which is sent via smartphone to the patient
and their doctor. Currently undergoing medical trials, to date the iT Bra has been
tested with 500 patients and has demonstrated an 87 percent correlation to a verified,
clinical diagnosis of breast cancer. In order for the cell temperature readings to be
taken, women wear the iT Bra for 12 hours and it is being presented as an alternative
to a monthly breast examination. Early detection and treatment has been shown to
improve breast cancer survival rates and Cyrcadia believes its technology could reduce
the number of unnecessary breast biopsies by up to 50 percent, especially among
women who have dense breast tissue, for whom mammography often doesn’t work.
A key feature of our proposed detection concept is the simultaneous sensing of
tissue property changes to the two female breasts since the right and left healthy
breasts are morphologically and materially identical. It’s developed a breast health
system made up of two parts. The first is a sensor device that’s placed in a bra to
measure cell temperature changes created over time by new blood vessel growth
associated with developing tumors. The second is proprietary software that uses
pattern recognition, chronology and artificial intelligence to look for changes in breast
tissue that could indicate the presence of a tumor. Doctors and researchers say the size
of breast cancer and how far it has spread are some of the most important factors in
determining the prognosis of the more than 225,000 women alone who are diagnosed

215
A Survey on Early Detection of Women’s Breast Cancer Using IoT

with invasive breast cancer each year. Recent advancements in breast imaging like
3D mammography and thermography may help in detection and diagnosis, but there
are still flaws in the gold standard of mammography (which have stirred up some
debate about its value as an annual screening method for women over 40).
The sensors are contained in a patch that attaches with an adhesive to the patient’s
skin. The patient wears the iTBra for 2 hours, and the data collected is sent directly
to her physician for analysis. It’s an alternative to the discomfort of a mammogram,
and it’s especially helpful for women with dense breast tissue, such as Royea’s own
wife, Kelli Royea, who is featured in the documentary.

Figure 2. Self Examination using IOT Sensors

Mammography is the current gold standard diagnostic tool for breast cancer
screening. But screening mammography has an important limitation: its results are
significantly less accurate in women with dense breast tissue. Breast tissue density is
a recognized medical condition which affects more than 40% of women worldwide.
Dense breast tissue is comprised of less fat and more connective/fibrous and glandular
tissue, and ranges in severity from Level A (fatty) to Level D (extremely dense).
As the density of a breast increases, the ability of the mammogram to reveal cancer
decreases. Because both dense breast tissue and breast cancer appear white on
mammography images, finding cancer in these dense tissue breasts is akin to looking
for a distinct snowflake in a snowstorm. The cancer risk in women with extremely
dense breasts is up to 6 times higher compared to normal/fatty tissue, and shows a
much more rapid acceleration of the condition. Still, 70% of breast biopsies that are
conducted as a result of suspicious findings on a mammogram are performed on

216
A Survey on Early Detection of Women’s Breast Cancer Using IoT

non-cancerous tissue. The staggering numbers of such unnecessary biopsies could

be reduced with improved diagnostic screening in women with dense breast tissue.
One recent development, however, has really captured my attention because of
its potential to help with early detection of breast cancer. That’s the idea behind the
connected bra—dubbed the” iTBra” by its inventor, Rob Royea, CEO of Cyrcadia
Health. With embedded temperature sensors, this new kind of wearable technology
tracks changes in temperature in breast tissue over time. It uses machine learning and
predictive analytics to identify and classify abnormal patterns that could indicate early
stage breast cancer. And women will be glad to know that they just need to wear the
iTBra for 2 to 12 hours once a month as they go about their daily activities—there’s
no painful squashing or prodding or radiation involved. Internet of Things (IoT)
across many industries – manufacturing, transportation, agriculture, smart cities,
retail, and finance. So many new solutions are on display that help organizations
either save or make money. But in healthcare, IoT can actually do more than that, it
has the potential to save lives. The Internet of Things (IoT) also has the potential to
bring a huge transformation to the Healthcare industry. For the starters, it promises
to reduce the emergency room wait time, track patient data accurately, and manage
healthcare inventory. All these functions will improve the efficiency of the healthcare
sector to an unimaginable extent.
The iTBra is comprised of two wearable breast sensor patches; flexible and
intelligent, they detect dynamic circadian temperature changes within breast tissue.
In contrast with mammography and ultrasound, which rely on specular reflective
capabilities of varying tissue densities, the Cyrcadia Health solution is tissue agnostic
and is able to detect early circadian cellular changes in all tissue types and varied
age groups. The abnormalities in the circadian rhythm-based temperature variances
of cell cycles are present at the earliest stages of abnormal cellular growth and
proliferation, and serve as the early indicators of breast cancer. Exclusively patented
predictive analytics technology identifies these abnormal cellular changes and reports
results to the health care provider, used to assist in their clinical decision process.

CONCLUSION

These methods are impractical to be used as personal monitoring device due to its
high cost and uncomfortable procedures on the patient. Therefore, this study proposes
the use of multiple sensors positioned on brassiere cloth covering all four quadrants
to provide continuous monitoring of temperature changes on the breasts. To test the
reliability of the developed device, breast phantom and heater were used to mimic
women breasts and the tumor respectively. Camera was used to verify the changes
of surface temperature on breast phantom. Result obtained shows that the reading

217
A Survey on Early Detection of Women’s Breast Cancer Using IoT

of sensors on each quadrant with a tumor to have higher temperature compared to

the rest. Its indicates that this low-cost wearable device can be potentially used as
breast cancer monitoring tool.

REFERENCES

Andreu-Perez, J., Leff, D. R., Ip, H. M. D., & Yang, G.-Z. (2015). From Wearable
Sensors to Smart Implants – Towards Pervasive and Personalised Healthcare.
IEEE Transactions on Biomedical Engineering, 62(12), 2750–2762. doi:10.1109/
TBME.2015.2422751 PMID:25879838
Arcarisi & Di Pietro. (2019). Palpreast—A New Wearable Device for Breast Self-
Examination. Appl. Sci., 9(3).
Atallah, L. (2011). Observing recovery from knee-replacement surgery by using
wearable sensors. IEEE Digital Xplore.
Kong, K., & Tomizuka, M. (2009). A gait monitoring system based on air pressure
sensors embedded in a shoe. IEEE/ASME Transactions on Mechatronics, 14(3),
358–370. doi:10.1109/TMECH.2008.2008803
Liana, D. D., Raguse, B., Gooding, J. J., & Chow, E. (2012). Recent Advances in
Paper-Based Sensors. Sensors (Basel), 12(9), 11505–11526. doi:10.3390120911505
PMID:23112667
Mannoor, M. S., Tao, H., Clayton, J. D., Sengupta, A., Kaplan, D. L., Naik, R. R.,
Verma, N., Omenetto, F. G., & McAlpine, M. C. (2012). Graphene-based wireless
bacteria detection on tooth enamel. Nature Communications, 3(1), 1–8. doi:10.1038/
ncomms1767 PMID:22453836
Memon & Li. (2019). Breast Cancer Detection in the IOT Health Environment
Using Modified Recursive Feature Selection. Wireless Communications and Mobile
Computing, 1-19.
Milon Islam, Md., & Rashedul Islam, Md. (2020). Development of Smart Healthcare
Monitoring System in IoT Environment. SN Computer Science, 185, 1–11.
Ng, K.-G., Wong, S.-T., Lim, S.-M., & Goh, Z. (2010). Evaluation of the cadi
thermosensor wireless skin-contact thermometer against ear and axillary temperatures
in children. Journal of Pediatric Nursing-Nursing Care of Children & Families,
25(3), 176–186. doi:10.1016/j.pedn.2008.12.002 PMID:20430278
O’Flynn, J. (2013). Novel smart sensor glove for arthritis rehabilitation. BSN, 1-6.

218
A Survey on Early Detection of Women’s Breast Cancer Using IoT

Patel, S., Lorincz, K., Hughes, R., Huggins, N., Growdon, J., Standaert, D., Akay,
M., Dy, J., Welsh, M., & Bonato, P. (2009). Monitoring motor fluctuations in patients
with parkinson’s disease using wearable sensors. IEEE Transactions on Information
Technology in Biomedicine, 13(6), 864–873. doi:10.1109/TITB.2009.2033471
PMID:19846382
Po, M.-Z., Swenson, C., & Rosalind, W. (2010). A Wearable Sensor for Unobtrusive,
Long-Term Assessment of Electrodermal Activity. IEEE Transactions on Biomedical
Engineering, 57(5), 1243–1252. doi:10.1109/TBME.2009.2038487 PMID:20172811
Sazonov, E. S., Fulk, G., Hill, J., Schutz, Y., & Browning, R. (2011). Monitoring of
posture allocations and activities by a shoe-based wearable sensor. IEEE Transactions
on Biomedical Engineering, 58(4), 983–990. doi:10.1109/TBME.2010.2046738
PMID:20403783
Sung, M., Marci, C., & Pentland, A. (2005). Carl Marci, Alex Pentland, “Wearable
feedback systems for rehabilitation. Journal of Neuroengineering and Rehabilitation,
2(1), 1–12. doi:10.1186/1743-0003-2-17
Xu, S., Zhang, Y., Jia, L., & Kyle, E. (2014). Soft microfluidic assemblies
of sensors, circuits, and radios for the Skin. Science, 344(6179), 70–74.
doi:10.1126cience.1250169 PMID:24700852
Zhou, H., Stone, T., Hu, H., & Harris, N. (2008). Use of multiple wearable inertial
sensors in upper limb motion tracking. Medical Engineering & Physics, 30(1),
123–133. doi:10.1016/j.medengphy.2006.11.010 PMID:17251049

219
220

Compilation of References

Aaron, J. S., Taylor, A. B., & Chew, T. L. (2018). Image co-localization–co-occurrence versus
correlation. Journal of Cell Science, 131(3), jcs211847. doi:10.1242/jcs.211847 PMID:29439158

Abdel-Zaher, A. M., & Eldeib, A. M. (2016). Breast cancer classification using deep belief
networks. Expert Systems with Applications, 46(1), 139–144. doi:10.1016/j.eswa.2015.10.015

Aggarwal, C. C. (2015). Outlier analysis. Data Mining, 237-263. doi:10.1007/978-3-319-14142-8_8

Aggarwal, R., & Kaur, A. (2012). Comparative Analysis of Different Algorithms For Brain
Tumor Detection. International Journal of Scientific Research.

Alharkan, T., & Martin, P. (2012). IDSaaS: Intrusion detection system as a service in public
clouds. Proceedings of the 2012 12th IEEE/ACM International Symposium on Cluster, Cloud
and Grid Computing (ccgrid 2012), 686-687. 10.1109/CCGrid.2012.81

Aloudat & Faezipour. (2016). Determination for Glaucoma Disease Based on Red Area Percentage.
2016 IEEE Long Island Systems, Applications and Technology Conference (LISAT).

American Cancer Society. (2019). Cancer Facts & Figures 2019. American Cancer Society.

Anthony, M., & Bartlet, P. (1999). Neural Network Learning: Theoretical Foundations. Cambridge
University Press. doi:10.1017/CBO9780511624216

Araujo, M., Queiroz, K., Pininga, M., Lima, R., & Santos, W. (2012). Uso de regiões elipsoidais
como ferramenta de segmentação em termogramas de mama. In XXIII Congresso Brasileiro de
Engenharia Biomédica (CBEB 2012). Pernambuco: SBEB.

Arcarisi & Di Pietro. (2019). Palpreast—A New Wearable Device for Breast Self-Examination.
Appl. Sci., 9(3).

Aslam, T. M., Tan, S. Z., & Dhillon, B. (2009). Iris recognition in the presence of ocular disease.
Journal of the Royal Society, Interface, 6(34), 2009. doi:10.1098/rsif.2008.0530 PMID:19324690

Atallah, L. (2011). Observing recovery from knee-replacement surgery by using wearable sensors.
IEEE Digital Xplore.

Azevedo, W. W., Lima, S. M., Fernandes, I. M., Rocha, A. D., Cordeiro, F. R., Silva-Filho, A.
G., & Santos, W. P. (2015). Fuzzy Morphological Extreme Learning Machines to Detect and
Classify Masses in Mammograms. In 2015 IEEE International Conference on Fuzzy Systems.
IEEE. 10.1109/FUZZ-IEEE.2015.7337975

Bace & Mell. (2001). NIST special publication on intrusion detection systems. Booz-Allen and
Hamilton Inc.

Bagahel, D., & Kiran, K.G. (2015). Comparative study of tumor detection techniques with their
suitability for brain MRI images. Inlt. Jrl., 127(13).

Balafar, M. A., Ramli, A. R., Saripan, M. I., & Mashohor, S. (2010). Review of brain MRI
segmentation methods. Artificial Intelligence Review, 33(3), 261–274. doi:10.100710462-010-
9155-0

Bandyopadhyay, S. K. (2010). Survey on Segmentation Methods for Locating Masses in a

Mammogram Image. International Journal of Computers and Applications, 9(11), 25–28.
doi:10.5120/1429-1926

Bashir, U., & Chachoo, M. (2014). Intrusion detection and prevention system: Challenges &
opportunities. Computing for Sustainable Global Development (INDIACom), 2014 International
Conference on, 806-809. 10.1109/IndiaCom.2014.6828073

Bayramoglu, N., Kannala, J., & Heikkila, J. (2016). Deep Learning for Magnification Independent
Breast Cancer Histopathology Image Classification. In 2016 23rd International Conference on
Pattern Recognition (ICPR). Cancun: IEEE. 10.1109/ICPR.2016.7900002

Benson, C. C. (2016). Brain Tumor Segmentation from MR Brain Images using Improved Fuzzy
c-Means Clustering and Watershed Algorithm. 2016 Intl. Conference on Advances in Computing,
Communications and Informatics (ICACCI). 10.1109/ICACCI.2016.7732045

Bharati, M., & Tamane, S. (2017). Intrusion detection systems (IDS) & future challenges in
cloud based environment. Intelligent Systems and Information Management (ICISIM), 240-250.
doi:10.1109/ICISIM.2017.8122180

221
Compilation of References

Bhateja, V., Gautam, A., Tiwari, A., Bao, L. N., Satapathy, S. C., Nhu, N. G., & Le, D.-N. (2018).
Haralick Features-Based Classification of Mammograms Using SVM. In Information Systems
Design and Intelligent Applications (Vol. 672). Springer. doi:10.1007/978-981-10-7512-4_77

Bhuvaneswari, Aruna, & Loganathan. (2014). Classification of Lung Diseases by Image Processing
Techniques Using Computed Tomography Images. International Journal of Advanced Computer
Research, 4(1).

Bjoern, H. M., & Koen, V. L. (2016). A Generative Probabilistic Model and Discriminative
Extensions for Brain Lesion Segmentation With Application to Tumor and Stroke. IEEE Transactions
on Medical Imaging, 35(4), 933–946. doi:10.1109/TMI.2015.2502596 PMID:26599702

Bolte, S., & Cordelières, F. P. (2006). A guided tour into subcellular colocalization analysis in
light microscopy. Journal of Microscopy, 224(3), 213–232. doi:10.1111/j.1365-2818.2006.01706.x
PMID:17210054

Borgen, H., Bours, P., & Wolthusen, S. D. (2009). Simulating the Influences of Aging and Ocular
Disease on Biometric Recognition Performance. International Conference on Biometrics 2009,
8(8), 857–867. 10.1007/978-3-642-01793-3_87

Budai, A., Bock, R., Maier, A., Hornegger, J., & Michelson, G. (2013). Robust Vessel Segmentation
in Fundus Images. International Journal of Biomedical Imaging.

Canadian Border Services Agency. (2015). CANPASS Air. Available: http: //www.cbsa-asfc.
gc.ca/prog/canpass/canpassair-eng.html

Candemir, Jaeger, Palaniappan, & Musco. (2014). Lung Segmentation in Chest Radiographs Using
Anatomical Atlases With Non-rigid Registration. IEEE Transactions on Medical Imaging, 33(2).

Chang, R. F., Wu, W. J., Moon, W. K., & Chen, D.-R. (2005). Dr Chen, “Automatic Ultrasound
Segmentation and Morphology based Diagnosis of Solid Breast Tumors. Breast Cancer Research
and Treatment, 89(2), 179–185. doi:10.100710549-004-2043-z PMID:15692761

Chang, S.-K., & Hsu, A. (1992). Image information systems: Where do we go from here? IEEE
Transactions on Knowledge and Data Engineering, 4(5), 431–442. doi:10.1109/69.166986

Chen, Y., Wang, Y., & Yang, B. (2006). Evolving Hierarchical RBF Neural Networks for Breast
Cancer Detection. LNCS, 4234, 137-144. doi:10.1007/11893295_16

Chung, C.-J., Khatkar, P., Xing, T., Lee, J., & Huang, D. (2013). NICE: Network Intrusion
Detection and Countermeasure Selection in Virtual Network Systems. IEEE Transactions on
Dependable and Secure Computing, 10(4), 198–211. doi:10.1109/TDSC.2013.8

222
Compilation of References

Commowick, O., Istace, A., Kain, M., Laurent, B., Leray, F., Simon, M., Pop, S. C., Girard, P.,
Ameli, R., Ferré, J.-C., Kerbrat, A., Tourdias, T., Cervenansky, F., Glatard, T., Beaumont, J.,
Doyle, S., Forbes, F., Knight, J., Khademi, A., ... Barillot, C. (2018). Objective Evaluation of
Multiple Sclerosis Lesion Segmentation Using a Data Management and Processing Infrastructure.
Scientific Reports, 8(1), 13650. doi:10.103841598-018-31911-7 PMID:30209345

Cordeiro, F. R., Bezerra, K. F. P., & Santos, W. P. (2017). Random walker with fuzzy initialization
applied to segment masses in mammography images. In 30th International Symposium on
Computer-Based Medical Systems (CBMS). IEEE. 10.1109/CBMS.2017.40

Cordeiro, F. R., Lima, S. M., Silva-Filho, A. G., & Santos, W. P. (2012). Segmentation of
mammography by applying extreme learning machine in tumor detection. In International
Conference of Intelligent Data Engineering and Automated Learning. Berlin: Springer.

Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2013). Segmentation of mammography by

applying growcut for mass detection. Studies in Health Technology and Informatics, 192, 87–91.
PMID:23920521

Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2016a). A semi-supervised fuzzy growcut
algorithm to segment and classify regions of interest of mammographic images. Expert Systems
with Applications, 65, 116–126. doi:10.1016/j.eswa.2016.08.016

Cordeiro, F. R., Santos, W. P., & Silva-Filho, A. G. (2016b). An adaptive semi-supervised fuzzy
growcut algorithm to segment masses of regions of interest of mammographic images. Applied
Soft Computing, 46, 613–628. doi:10.1016/j.asoc.2015.11.040

Corliss, B. A., Ray, H. C., Patrie, J. T., Mansour, J., Kesting, S., Park, J. H., & Peirce, S. M.
(2019). CIRCOAST: A statistical hypothesis test for cellular colocalization with network
structures. Bioinformatics (Oxford, England), 35(3), 506–514. doi:10.1093/bioinformatics/
bty638 PMID:30032263

Corso, J. J., Sharon, E., Dube, S., El-Saden, S., Sinha, U., & Yuille, A. (2008, May). Efficient
Multilevel Brain Tumor Segmentation with Integrated Bayesian Model Classification. IEEE
Transactions on Medical Imaging, 27(5), 629–640. doi:10.1109/TMI.2007.912817 PMID:18450536

Cruz, T. N., Cruz, T. M., & Santos, W. P. (2018). Detection and classification of lesions in
mammographies using neural networks and morphological wavelets. IEEE Latin America
Transactions, 16(3), 926–932. doi:10.1109/TLA.2018.8358675

223
Compilation of References

D’Orsi, C. J., Sickles, E. A., Mendelson, E. B., & Morris, E. A. (2013). Breast Imaging Reporting
and Data System: ACR BI-RADS breast imaging atlas (5th ed.). American College of Radiology.

Dai, S., Lu, K., & Dong, J. (2015). Lung segmentation with improved graph cuts on chest CT
images. 3rd IAPR Asian Conference on Pattern Recognition. 10.1109/ACPR.2015.7486502

Dean Bidgood, W. Jr. (1998). The SNOMED DICOM Microglossary: Controlled terminology
resource for Idata interchange in biomedical imaging. Methods of Information in Medicine,
37(4/5), 404–414. doi:10.1055-0038-1634557 PMID:9865038

Deserno, T., Soiron, M., Oliveira, J., & Araújo, A. (2012a). Towards Computer-Aided Diagnostics
of Screening Mammography Using Content-Based Image Retrieval. In 2011 24th SIBGRAPI
Conference on Graphics, Patterns and Images. Alagoas: IEEE.

Deserno, T. M., Soiron, M., Oliveira, J. E. E., & Araújo, A. A. (2012b). Computer-aided diagnostics
of screening mammography using content-based image retrieval. In Medical Imaging: Computer-
Aided Diagnosis 2012. SPIE. doi:10.1117/12.912392

Dhir, L., Habib, N. E., Monro, D. M., & Rakshit, S. (2010). Effect of cataract surgery and pupil
dilation on iris pattern recognition for personal authentication. Eye (London, England), 24(6),
1006–1010. doi:10.1038/eye.2009.275 PMID:19911017

Dhooge, M., & de Laey, J. J. (1989). The ocular ischemic syndrome. Bulletin de la Société Belge
d’Ophtalmologie, 231, 1–13. PMID:2488440

Duan, Y., Wang, J., Hu, M., Zhou, M., Li, Q., Sun, L., & Wang, Y. (2019). Leukocyte classification
based on spatial and spectral features of microscopic hyperspectral images. Optics & Laser
Technology, 112, 530–538. doi:10.1016/j.optlastec.2018.11.057

Durgadevi & Shekhar. (2015). Identification of tumor using K-means algorithm. Intl. Jrl. Adv.
Res. Inn. Id. Edu, 1, 227-231.

Dvorak, P., & Menze, B. (2015). Structured prediction with convolutional neural networks
for multimodal brain tumor segmentation. Proceeding of the Multimodal Brain Tumor Image
Segmentation Challenge, 13-24.

Elbalaoui, Fakir, Taifi, & Merbohua. (2016). Automatic Detection of Blood Vessel in Retinal
Images. 13th International Conference Computer Graphics, Imaging and Visualization.

Eltoukhy, M., Faye, I., & Samir, B. (2009). Breast Cancer Diagnosis in Mammograms using
Multilevel Wavelet Analysis. Proceeding of National Postgraduate Conference.

224
Compilation of References

Eman, A. M., Mohammed, E., & Rashid, A. L. (2015). Brain tumor segmentation based on a hybrid
clustering technique. Egyptian Informatics Journal, 16(1), 71–81. doi:10.1016/j.eij.2015.01.003

Emran, Abtin, & David. (2015). Automatic segmentation of wrist bones in CT using a statistical
wrist shape pose model. Inlt. Jrl.

Fathy, M., Keshk, M., & El Sherif, A. (2019). Surgical management and outcome of intramedullary
spinal cord tumour. Egyptian Journal of Neurosurgery., 34(2), 2–7. doi:10.118641984-019-0028-9

Fernandes, I., & Santos, W. (2014). Classificação de mamografias utilizando extração de atributos
de textura e redes neurais artificiais. In Congresso Brasileiro de Engenharia Biomédica (CBEB
2014). SBEB.

Ferreira, J., Oliveira, H., & Martinez, M. (2011). Aplicação de uma metodologia computacional
inteligente no diagnóstico de lesões cancerígenas. Revista Brasileira de Inovação Tecnológica
em Saúde, 1(2), 4-9.

Freixenet, J., Munoz, X., Raba, D., Marti, J., & Cufi, X. (2002). Yet another survey on image
segmentation: Region and boundary information integration. Proc. 7th Eur. Conf. Computer
Vision Part III, 408–422. 10.1007/3-540-47977-5_27

Fuadah, Setiawan, & Mengko. (2015). Mobile Cataract Detection using Optimal Combination of
Statistical Texture Analysis. 4th International Conference on Instrumentation, Communications,
Information Technology, and Biomedical Engineering (ICICI-BME).

Fuadah, Setiawan, Mengko, & Budiman. (2015). A computer aided healthcare system for cataract
classification and grading based on fundus image analysis. Elsevier Science Publishers B. V.

Ganesan, K., Acharya, U. R., Chua, C. K., Min, L. C., & Abraham, T. K. (2014). Automated
Diagnosis of Mammogram Images of Breast Cancer Using Discrete Wavelet Transform and
Spherical Wavelet Transform Features: A Comparative Study. Technology in Cancer Research
& Treatment, 13(6), 605–615. doi:10.7785/tcrtexpress.2013.600262 PMID:24000991

Georges, B. A. (1999). Model Creation and Deformation for the Automatic Segmentation of
the Brain in MR Images. IEEE Transactions on Biomedical Engineering, 46(11), 1346–1356.
doi:10.1109/10.797995 PMID:10582420

Girisha, Chandrashekhar, & Kurian. (2013). Texture Feature Extraction of Video Frames Using
GLCM. International Journal of Engineering Trends and Technology, 4(6).

Girshick, R. (2014). Rich feature hierarchies for accurate object detection and semantic
segmentation. Proceedings of the IEEE conference on computer vision and pattern recognition.
10.1109/CVPR.2014.81

Gogic, I., Manhart, M., Pandzic, I. S., & Ahlberg, J. (2020). Fast facial expression recognition
using local binary features and shallow neural networks. The Visual Computer, 36(1), 97–112.
doi:10.100700371-018-1585-8

225
Compilation of References

Guld, Kohnen, Keysers, Schubert, Wein, Bredno, & Lehmann. (2002). Quality of DICOM header
information for image categorization. SER, Proc. SPIE, 4685, 280-287. 10.1117/12.467017

Guo, Z., Zhang, L., & Zhang, D. (2010). Rotation invariant texture classification using LBP variance
with global matching. Pattern Recognition, 43(3), 706–716. doi:10.1016/j.patcog.2009.08.017

Gupta & Singh. (2017). Brain Tumor segmentation and classification using Fcm and support
vector machine. International Research Journal of Engineering and Technology, 4(5).

Hai, S., Xing, F., & Yang, L. (2016). Robust Cell Detection of Histopathological Brain Tumor
Images Using Sparse Reconstruction and Adaptive Dictionary Selection. IEEE Transactions on
Medical Imaging, 35(6), 1575–1586. doi:10.1109/TMI.2016.2520502 PMID:26812706

Haleem, M. S., Han, L., van Hemert, J., & Fleming, A. (2015). Glaucoma Classification using
Regional Wavelet Features of the ONH and its Surroundinga. 37th Annual International Conference
of the IEEE Engineering in Medicine and Biology Society (EMBC).

Haralick, R. M., Shanmugam, K., & Dinstein, I. (1973). Textural Features for Image
Classification. IEEE Transactions on Systems, Man, and Cybernetics, 3(6), 610–621. doi:10.1109/
TSMC.1973.4309314

Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C., Jodoin,
P.-M., & Larochelle, H. (2017). Brain tumor segmentation with deep neural networks. Medical
Image Analysis, 35, 18–31. doi:10.1016/j.media.2016.05.004 PMID:27310171

Hazlina, H., & Sameem, A. K. (2004). Back Propagation Neural Network for the Prognosis of
Breast Cancer: Comparison on Different Training Algorithms. Proceedings Second International
Conference on Artificial Intelligence in Engineering & Technology, 445-449.

Heath, M., Bowyer, K. W., & Kopans, D. (2000). The digital database for screening mammography.
Proceedings of the 5th International Workshop on Digital Mammography

Hersh, W., Muller, H., & Kalpathy. (2009). The imageCLEFmed medical image retrieval task
test collection. J. Digital Imaging, 22(6), 648-655.

Hogeweg, Sánchez, Maduskar, Philipsen, & Story. (2015). Automatic Detection of Tuberculosis
in Chest Radiographs Using a Combination of Textural, Focal and Shape Abnormality Analysis.
IEEE Transactions on Medical Imaging, 34(12).

Horsch, K., Giger, M. L., Venkata, L. A., & Vybomya, C. J. (2001). Automatic Segmentation
of Breast Lesions on Ultrasound. Medical Physics, 28(8), 1652–1659. doi:10.1118/1.1386426
PMID:11548934

Hotko, Y. S. (2013). Male breast cancer: Clinical presentation, diagnosis, treatment. Experimental
Oncology, 35(4), 303–310. PMID:24382442

Huang, H. K. (2004). PACS and imaging informatics: basic principles and applications. John
Wiley & Sons Inc. doi:10.1002/0471654787

226
Compilation of References

Hussain, C. A., Rao, D. V., & Mastani, S. A. (2020). RetrieveNet: a novel deep network for
medical image retrieval. Evol. Intel. doi:10.100712065-020-00401-z

ISO/IEC 19794-6:2011. (2011). Information technology – Biometric data interchange formats

– Part 6: Iris image data.

Jaeger, Karargyris, Candemir, Folio, & Siegelman. (2014). Automatic Tuberculosis Screening
Using Chest Radiographs. IEEE Transactions on Medical Imaging, 33(2).

Jagath, C. (2001, October). Bayesian Approach to Segmentation of Statistical Parametric Maps.

IEEE Transactions on Biomedical Engineering, 48(10).

Jannesari, M., Habibzadeh, M., Aboulkheyr, H., Khosravi, P., Elemento, O., Totonchi, M., &
Hajirasouliha, I. (2018). Breast Cancer Histopathological Image Classification: A Deep Learning
Approach. In 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM).
IEEE. 10.1109/BIBM.2018.8621307

Jobin Christ, M. C., Sasikumar, K., & Parwathy, R. M. S. (2009, July). Application of Bayesian
Method in Medical Image Segmentation. International Journal of Computing Science and
Communication Technologies, VOL, 2(1).

Jose, Ravi, & Sampath. (2014). Brain tumor segmentation a performance analysis using K-means,
fuzzy c-means and region growing algorithm. Inlt. Jrl., 2(3).

Joseph, S., & Balakrishnan, K. (2011). Local Binary Patterns, Haar Wavelet Features and Haralick
Texture Features for Mammogram Image Classification using Artificial Neural Networks. In
International Conference on Advances in Computing and Information Technology. Springer.
10.1007/978-3-642-22555-0_12

Juang, L. H., & Ming, N. W. (2010). MRI brain lesion image detection based on color-converted
K-means. Science Direct Measurement, 43(7), 941–949.

Juhl, J. H., Crummy, A. B., & Kuhlman, J. E. (2000). Paul & Juhl Interpretação Radiológica (7a
ed.). Rio de Janeiro: Guanabara-Koogan.

Kamnitsas, K., Ledig, C., Newcombe, V. F. J., Simpson, J. P., Kane, A. D., Menon, D. K., Rueckert,
D., & Glocker, B. (2017). Efficient multi-scale 3D CNN with fully connected CRF for accurate
brain lesion segmentation. Medical Image Analysis, 36, 61–78. doi:10.1016/j.media.2016.10.004
PMID:27865153

Kaur & Rani. (2016). MRI Brain Tumor Segmentation Methods- A Review. International Journal
of Current Engineering and Technology.

Khadem. (2010). MRI Brain image segmentation using graph cuts (Master’s thesis). Chalmers
University of Technology, Goteborg, Sweden.

227
Compilation of References

Kholidy, H. A., & Baiardi, F. (2012). CIDS: A framework for intrusion detection in cloud systems.
Information Technology: New Generations (ITNG), Ninth International Conference on, 379-385.
10.1109/ITNG.2012.94

Khuriwal, N., & Mishra, N. (2018). Breast Cancer Detection from Histopathological Images using
Deep Learning. In 2018 3rd International Conference and Workshops on Recent Advances and
Innovations in Engineering (ICRAIE). IEEE. 10.1109/ICRAIE.2018.8710426

Kishore, B., Arjunan, R. V., Saha, R., & Selvan, S. (2014). Using Haralick Features for the
Distance Measure Classification of Digital Mammograms. International Journal of Computers
and Applications, 6(1), 17–21.

Kiyan, T., & Yildrim, T. (2004). Breast Cancer Diagnosis using Statistical Neural Networks.
Journal of Electrical and Electronics Engineering (Oradea), 4(2), 1149–1153.

Kong, K., & Tomizuka, M. (2009). A gait monitoring system based on air pressure sensors
embedded in a shoe. IEEE/ASME Transactions on Mechatronics, 14(3), 358–370. doi:10.1109/
TMECH.2008.2008803

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep
convolutional neural networks. Advances in Neural Information Processing Systems.

Krupinski. (2000). The Importance of Perception Research in Medical Imaging. Radiation

Medicine, 8(6), 329-334.

Kumar, Manjunathand, & Sheshadri. (2015). Feature extraction from the fundus images for the
diagnosis of diabetic retinopathy. International Conference on Emerging Research in Electronics.
Computer Science and Technology.

Kumar, A. (2017). A Novel Approach for Brain Tumor Detection Using Support Vector Machine.
K-Means and PCA Algorithm.

Kumar, D. (2018). A modified intuitionistic fuzzy c-means clustering approach to segment human
brain MRI image. Multimedia Tools and Applications, 1–25.

Laddha, R. R. (2014). A Review on Brain Tumor Detection Using Segmentation And Threshold
Operations. International Journal of Computer Science and Information Technologies, 5(1),
607–611.

Langley, P. (1996). Elements of Machine Learning. Morgan Kaufmann.

Lavancier, F., Pécot, T., Zengzhen, L., & Kervrann, C. (2020). Testing independence between
two random sets for the analysis of colocalization in bioimaging. Biometrics, 76(1), 36–46.
doi:10.1111/biom.13115 PMID:31271216

Lee, J.-H., Park, M.-W., Eom, J.-H., & Chung, T.-M. (2011). Multi-level intrusion detection system
and log management in cloud computing. Advanced Communication Technology (ICACT), 13th
International Conference on, 552-555.

228
Compilation of References

Leemput, K. V., Maes, F., Vandermeulen, D., & Suetens, P. (1999). Automated model-based
tissue classification of MR images of brain. IEEE Transactions on Medical Imaging, 18(10),
897–908. doi:10.1109/42.811270 PMID:10628949

Lehmann, T. M., Guld, M. O., Thies, C., Fischer, B., Spitzer, K., Keysers, D., Ney, H., Kohnen,
M., Schubert, H., & Wein, B. B. (2004). Content-based image retrieval in medical applications.
Methods of Information in Medicine, 43(4), 354–361. doi:10.1055-0038-1633877 PMID:15472746

Liana, D. D., Raguse, B., Gooding, J. J., & Chow, E. (2012). Recent Advances in Paper-Based
Sensors. Sensors (Basel), 12(9), 11505–11526. doi:10.3390120911505 PMID:23112667

Liao, H.-J., Lin, C.-H. R., Lin, Y.-C., & Tung, K.-Y. (2013). Intrusion detection system: A
comprehensive review. Journal of Network and Computer Applications, 36(1), 16–24. doi:10.1016/j.
jnca.2012.09.004

Lima, S. M., Silva-Filho, A. G., & Santos, W. P. (2014). A methodology for classification of
lesions in mammographies using Zernike moments, ELM and SVM neural networks in a multi-
kernel approach. In 2014 IEEE International Conference on Systems, Man, and Cybernetics
(SMC). IEEE. 10.1109/SMC.2014.6974041

Linguraru, Richbourg, Liu, & Watt. (n.d.). Tumor Burden Analysis on Computed Tomography
by Automated Liver and Tumor Segmentation. IEEE.

Liu, H., Cao, H., & Song, E. (2019). Bone marrow cells detection: A technique for the
microscopic image analysis. Journal of Medical Systems, 43(4), 82. doi:10.100710916-019-
1185-9 PMID:30798374

Liu, Z. (2015). Semantic image segmentation via deep parsing network. Proceedings of the IEEE
International Conference on Computer Vision. 10.1109/ICCV.2015.162

Logeswari & Karnan. (2010). An improved implementation of brain tumor detection using
segmentation based on soft computing. Journal of Cancer Research and Experimental Oncology,
2(1).

Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic
segmentation. Proceedings of the IEEE conference on computer vision and pattern recognition.

Long, L. R., Antani, S., Deserno, T. M., & Thoma, G. R. (2009). Content-Based Image Retrieval in
Medicine: Retrospective Assessment, State of the Art, and Future Directions. International Journal
of Healthcare Information Systems and Informatics, 4(1), 1–16. doi:10.4018/jhisi.2009010101
PMID:20523757

Lotankar, M., Noronha, K., & Koti, J. (2015). Detection of Optic Disc and Cup from Color
Retinal Images for Automated Diagnosis of Glaucoma. IEEE UP Section Conference on Electrical
Computer and Electronics (UPCON).

Madabhushi, A., & Metaxas, D. (2003). Combining low-, high-level and Empirical Domain
Knowledge for Automated Segmentation of Ultrasonic Breast Lesions. IEEE Transactions on
Medical Imaging, 22(2), 155–169. doi:10.1109/TMI.2002.808364 PMID:12715992

229
Compilation of References

Madhukumar, S., & Santhiyakumari, N. (2015). Evaluation of k-Means and fuzzy C-means
segmentation on MR images of brain. Egyptian Society of Radiology and Nuclear Medicine,
46(2), 475–479. doi:10.1016/j.ejrnm.2015.02.008

Maitra, I. K., Nag, S., & Bandyopadhyay, S. K. (2011). Identification of Abnormal Masses in
Digital Mammography Images. International Journal of Computer Graphics, 2(1).

Mallat, S. (1999). A Wavelet Tour of Signal Processing. Academic.

Malvia, S., Bagadi, S. A., Dubey, U. S., & Saxena, S. (2017). Epidemiology ofbreast cancer
in Indian women. Asia Pacific Journal of Clinical Oncology, 13(4), 289–295. doi:10.1111/
ajco.12661 PMID:28181405

Manders, E. M. M., Verbeek, F. J., & Aten, J. A. (1993). Measurement of co-localization

of objects in dual colour confocal images. Journal of Microscopy, 169(3), 375–382.
doi:10.1111/j.1365-2818.1993.tb03313.x

Manjunath, K. N., Renuka, A., & Niranjan, U. C. (2007). Linear models of cumulative distribution
function for content-based medical image retrieval. Journal of Medical Systems, 31(6), 433–443.
doi:10.100710916-007-9075-y PMID:18041275

Mannoor, M. S., Tao, H., Clayton, J. D., Sengupta, A., Kaplan, D. L., Naik, R. R., Verma, N.,
Omenetto, F. G., & McAlpine, M. C. (2012). Graphene-based wireless bacteria detection on
tooth enamel. Nature Communications, 3(1), 1–8. doi:10.1038/ncomms1767 PMID:22453836

Maria, J., Amaro, J., Falcao, G., & Alexandre, L. A. (2016). Stacked Autoencoders Using
Low-Power Accelerated Architectures for Object Recognition in Autonomous Systems. Neural
Processing Letters, 43(2), 445–458. doi:10.100711063-015-9430-9

Martin, D., Fowlkes, C., Tal, D., & Malik, J. (2001). A database of human segmented natural
images and its application to evaluating segmentationalgorithms and measuring ecological
statistics. Proc. 8th Int. Conf. Computer Vision, 2, 416–423. 10.1109/ICCV.2001.937655

Mascaro, A. A., Mello, C. A., Santos, W. P., & Cavalcanti, G. D. (2009). Mammographic images
segmentation using texture descriptors. In 2009 Annual International Conference of the IEEE
Engineering in Medicine and Biology Society. IEEE. 10.1109/IEMBS.2009.5333696

Masdari, M., & Ahmadzadeh, S. (2017). A survey and taxonomy of the authentication schemes
in Telecare Medicine Information Systems. Journal of Network and Computer Applications, 87,
1–19. doi:10.1016/j.jnca.2017.03.003

MathWorks. (2019). Deep Learning Toolbox ™ Reference. Author.

McConnon, G., Deravi, F., Hoque, S., Sirlantzis, K., & Howells, G. (2012). Impact of Common
Ophthalmic Disorders on Iris Recognition. 2012 5th IAPR International Conference on Biometrics
Compendium, 277–282.

230
Compilation of References

Mehmood, Y., Shibli, M. A., Habiba, U., & Masood, R. (2013). Intrusion detection system in
cloud computing: challenges and opportunities. 2013 2nd National Conference on Information
Assurance (NCIA), 59-66. 10.1109/NCIA.2013.6725325

Mell & Grance. (2011). The NIST definition of cloud computing. Academic Press.

Memon & Li. (2019). Breast Cancer Detection in the IOT Health Environment Using Modified
Recursive Feature Selection. Wireless Communications and Mobile Computing, 1-19.

Mengqiao, W., Jie, Y., Yilei, C., & Hao, W. (2017). The multimodal brain tumor image segmentation
based on convolutional neural networks. 2017 2nd IEEE International Conference on Computational
Intelligence and Applications (ICCIA), 336-339. 10.1109/CIAPP.2017.8167234

Miao, H., & Xiao, C. (2018). Simultaneous segmentation of leukocyte and erythrocyte in
microscopic images using a marker-controlled watershed algorithm. Computational and
Mathematical Methods in Medicine, 2018, 2018. doi:10.1155/2018/7235795 PMID:29681997

Milon Islam, Md., & Rashedul Islam, Md. (2020). Development of Smart Healthcare Monitoring
System in IoT Environment. SN Computer Science, 185, 1–11.

Modi, C. N., Patel, D. R., Patel, A., & Muttukrishnan, R. (2012). Bayesian Classifier and Snort
based network intrusion detection system in cloud computing. In Computing Communication &
Networking Technologies (pp. 1–7). ICCCNT. doi:10.1109/ICCCNT.2012.6396086

Modi, C. N., Patel, D. R., Patel, A., & Muttukrishnan, R. (2012). Bayesian Classifier and Snort
based network intrusion detection system in cloud computing. Third International Conference
on Computing Communication & Networking Technologies (ICCCNT), 1-7.

Modi, C., Patel, D., Borisaniya, B., Patel, H., Patel, A., & Rajarajan, M. (2013). A survey of
intrusion detection techniques in cloud. Journal of Network and Computer Applications, 36(1),
42–57. doi:10.1016/j.jnca.2012.05.003

Moles Lopez, X., Barbot, P., Van Eycke, Y. R., Verset, L., Trépant, A. L., Larbanoix, L., &
Decaestecker, C. (2015). Registration of whole immunohistochemical slide images: An efficient
way to characterize biomarker colocalization. Journal of the American Medical Informatics
Association: JAMIA, 22(1), 86–99. doi:10.1136/amiajnl-2014-002710 PMID:25125687

Monro, D. M., Rakshit, S., & Zhang, D. (2009). DCT-Based Iris Recognition. IEEE Transactions
on Pattern Analysis and Machine Intelligence, 29(4), 586–595. doi:10.1109/TPAMI.2007.1002
PMID:17299216

Moon, N., Bullitt, E., Leemput, K., & Gerig, G. (2002). Model based brain and tumor segmentation.
Int. Conf. on Pattern Recognition, 528-531. 10.1109/ICPR.2002.1044787

231
Compilation of References

Mooney. (2018 April). Blood cell images Version 6. Retrieved May 23 2020 from https://www.
kaggle.com/paultimothymooney/blood-cells

Muller, A.C., & Guido, S. (n.d.). Introduction to Machine Learning with Python. O’Reilly.

Muller, Michoux, Bandon, & Geissbuhler. (2007). A review of content-based image retrieval
systems in medical applications—clinical benefits and future directions. Intl. Journal of Medical
Informatics, 73(1), 1–23.

Murathoti Varshini, Barjo, & Tigga. (2020). Spine Magnetic Resonance Image Segmentation
Using Deep Learning Techniques. 2020 6th International Conference on Advanced Computing
and Communication Systems (ICACCS).

Nanni, L., Brahnam, S., & Lumini, A. (2012). A simple method for improving local binary patterns
by considering non-uniform patterns. Pattern Recognition, 45(10), 3844–3852. doi:10.1016/j.
patcog.2012.04.007

Naveen Kumar, B., Chauhan, R. P., & Dahiya, N. (2016). Detection of Glaucoma using Image
processing techniques: A Review. 2016 International Conference on Microelectronics, Computing
and Communications (MicroCom).

Neurotechnology. (2012). VeriEye SDK, v. 4.3. Available: https://www.neurotechnology.com/

verieye.html

Ng, K.-G., Wong, S.-T., Lim, S.-M., & Goh, Z. (2010). Evaluation of the cadi thermosensor
wireless skin-contact thermometer against ear and axillary temperatures in children. Journal
of Pediatric Nursing-Nursing Care of Children & Families, 25(3), 176–186. doi:10.1016/j.
pedn.2008.12.002 PMID:20430278

Nikolai, J., & Wang, Y. (2014). Hypervisor-based cloud intrusion detection system. In Computing
(pp. 989–993). Networking and Communications.

Niwas, Lin, Kwoh, Kuo, Sng, Aquino, & Chew. (2016). Cross-examination for Angle-Closure
Glaucoma Feature Detection. IEEE Journal of Biomedical and Health Informatics.

Odstrcilik, J., Budai, A., Kolar, R., & Hornegger, J. (2013, June). Retinal vessel segmentation
by improved matched filtering: Evaluation on a new high-resolution fundus image database. IET
Image Processing, 7(4), 373–383. doi:10.1049/iet-ipr.2012.0455

O’Flynn, J. (2013). Novel smart sensor glove for arthritis rehabilitation. BSN, 1-6.

Ojala, T., Pietikainen, M., & Harwood, D. (1996). A comparative study of texture measures with
classification based on feature distributions. Pattern Recognition, 29(1), 51–59. doi:10.1016/0031-
3203(95)00067-4

Oktay, U., & Sahingoz, O. K. (2013). Proxy network intrusion detection system for cloud
computing. International Conference on Technological Advances in Electrical, Electronics and
Computer Engineering (TAEECE), 98-104. 10.1109/TAEECE.2013.6557203

232
Compilation of References

Omisore. (2014). Proposed the genetic neuro-fuzzy inferential model for the diagnosis of
tuberculosis. IEEE Transactions.

Onal, Susana, Paul, & Alferedo. (2014). Automated localization of multiple pelvic bone structure
in MRI. Intl Jrl.

Papakostas, G. A., Koulouriotis, D. E., Karakasis, E. G., & Tourassis, V. D. (2013). Moment-
based local binary patterns: A novel descriptor for invariant pattern recognition applications.
Neurocomputing, 99, 358–371. doi:10.1016/j.neucom.2012.06.031

Patel, A., Taghavi, M., Bakhtiyari, K., & Celestino Júnior, J. (2013). An intrusion detection and
prevention system in cloud computing: A systematic review. Journal of Network and Computer
Applications, 36(1), 25–41. doi:10.1016/j.jnca.2012.08.007

Patil. (2005). Pachpande, Automatic Brain Tumor Detection Using K-Means. Academic Press.

Pereira, Fonseca-Pinto, Paiva, Tavora, Assuncao, & Faria (2020). Accurate segmentation of
desmoscopic images based on local binary pattern clustering. International Convention on
Information and Communication Technology, Electronics and Microelectronics, Opatija, Croatia.

Pereira, S., Pinto, A., Alves, V., & Silva, C. A. (2016). Brain tumor segmentation using convolutional
neural networks in MRI images. IEEE Transactions on Medical Imaging, 35(5), 1240–1251.
doi:10.1109/TMI.2016.2538465 PMID:26960222

Pham, T. X., Siarry, P., & Oulhadj, H. (2018). Integrating fuzzy entropy clustering with an
improved PSO for MRI brain image segmentation. Applied Soft Computing, 65, 230–242.
doi:10.1016/j.asoc.2018.01.003

Po, M.-Z., Swenson, C., & Rosalind, W. (2010). A Wearable Sensor for Unobtrusive, Long-Term
Assessment of Electrodermal Activity. IEEE Transactions on Biomedical Engineering, 57(5),
1243–1252. doi:10.1109/TBME.2009.2038487 PMID:20172811

Pradhan, S. (2010). Development of Unsupervised Image Segmentation Schemes for Brain MRI
using HMRF model (Master Thesis). Department of EE, NIT, Rourkela, India.

233
Compilation of References

Priya. (2018). Efficient fuzzy c-means based multilevel image segmentation for brain tumor
detection in MR images. Design Automation for Embedded Systems, 1–13.

Putzu, L., Caocci, G., & Di Ruberto, C. (2014). Leucocyte classification for leukaemia detection
using image processing techniques. Artificial Intelligence in Medicine, 62(3), 179–191.
doi:10.1016/j.artmed.2014.09.002 PMID:25241903

Raad, A., Kalakech, A., & Ayache, M. (2012). Breast Cancer Classification using Neural Network
Approach: MLP AND RBF. The 13th international Arab conference on information technology,
10 – 13.

Rajendra Acharya, U. (2011, May). Automated Diagnosis of Glaucoma Using Texture and Higher
Order Spectra Features. IEEE Transactions on Information Technology in Biomedicine, 15(3).

Rendon-Gonzalez & Ponomaryov. (2016). Automatic Lung Nodule Segmentation and Classification
in CT Images Based on SVM. International conferences IEEE.

Roberts, T., Newell, M., Auffermann, W., & Vidakovic, B. (2017). Wavelet-based scaling indices
for breast cancer diagnostics. Statistics in Medicine, 36(12), 1989–2000. doi:10.1002im.7264
PMID:28226399

Roizenblatt, R., Schor, P., Dante, F., Roizenblatt, J., & Jr, R. B. (2004). Iris recognition as a
biometric method after cataract surgery. BioMedical Engineering Online, 3(2). www.biomedical-
engineering-online/com/content/3/1/2

Sachdeva, & Singh. (2015). Automatic Segmentation and Area Calculation of Optic Disc in
Ophthalmic Images. 2nd International Conference on Recent Advances in Engineering &
Computational Sciences (RAECS).

Sahiner, B., Heang-Ping, C., Patrick, N., Wei, D. M. A., Helie, D., Adler, D., & Goodsitt, M.
M. (1996). Classification of Mass and Normal Breast Tissue: A Convolution Neural Network
Classifier with Spatial Domain and Texture Images. IEEE Transactions on Medical Imaging,
15(5), 598–610. doi:10.1109/42.538937 PMID:18215941

Sahlol, A. T., Kollmannsberger, P., & Ewees, A. A. (2020). Efficient classification of white blood
cell leukemia with improved Swarm optimization of deep features. Scientific Reports, 10(1),
1–11. doi:10.103841598-020-59215-9 PMID:32054876

Salam, Akram, Abbas, & Anwar. (2015). Optic Disc Localization using Local Vessel Based
Features and Support Vector Machine. IEEE 15th International Conference on Bioinformatics
and Bioengineering (BIBE).

Santana, M. A., Pereira, J. M. S., Silva, F. L., Lima, N. M., Sousa, F. N., Arruda, G. M. S., Lima,
R. C. F., Silva, W. W. A., & Santos, W. P. (2018). Breast cancer diagnosis based on mammary
thermography and extreme learning machines. Research on Biomedical Engineering, 34(1),
45–53. doi:10.1590/2446-4740.05217

234
Compilation of References

Santana, M., Pereira, J., Lima, N., Sousa, F., Lima, R., & Santos, W. (2017). Classificação de
lesões em imagens frontais de termografia de mama a partir de sistema inteligente de suporte ao
diagnóstico. In Anais do I Simpósio de Inovação em Engenharia Biomédica. SABIO.

Santos, W. P., Assis, F. M., Souza, R. E., Mendes, P. B., Monteiro, H. S. S., & Alves, H. D. (2009a).
A Dialectical Method to Classify Alzheimer’s Magnetic Resonance Images. In Evolutionary
Computation. IntechOpen. doi:10.5772/9609

Santos, W. P., Assis, F. M., Souza, R. E., Mendes, P. B., Monteiro, H. S. S., & Alves, H. D. (2010).
Fuzzy-based Dialectical Non-supervised Image Classification and Clustering. International
Journal of Hybrid Intelligent Systems, 7(2), 115–124. doi:10.3233/HIS-2010-0108

Santos, W. P., Assis, F. M., Souza, R. E., & Santos Filho, P. B. (2008a). Evaluation of Alzheimer’s
Disease by Analysis of MR Images using Objective Dialectical Classifiers as an Alternative to
ADC Maps. In 2008 30th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society. IEEE.

Santos, W. P., Assis, F. M., Souza, R. E., & Santos Filho, P. B. (2009). Dialectical Classification
of MR Images for the Evaluation of Alzheimer’s Disease. In Recent Advances in Biomedical
Engineering. IntechOpen. doi:10.5772/7475

Santos, W. P., Assis, F., Souza, R., Santos Filho, P. B., & Neto, F. L. (2009b). Dialectical Multispectral
Classification of Diffusion-weighted Magnetic Resonance Images as an Alternative to Apparent
Diffusion Coefficients Maps to Perform Anatomical Analysis. Computerized Medical Imaging
and Graphics, 33(6), 442–460. doi:10.1016/j.compmedimag.2009.04.004 PMID:19446434

Santos, W. P., Souza, R. E., Santos Filho, P. B., Neto, F. B. L., & Assis, F. M. (2008b). A
Dialectical Approach for Classification of DW-MR Alzheimer’s Images. In 2008 IEEE Congress
on Evolutionary Computation (IEEE World Congress on COmputational Intelligence). Hong
Kong: IEEE. 10.1109/CEC.2008.4631023

Sapra, P., Singh, R., & Khurana, S. (2013). Brain Tumor Detection using Neural Network.
International Journal of Science and Modern Engineering, 1(9).

Sathies Kumar, T. (2017). Brain Tumor Detection Using SVM Classifier. IEEE 3rd International
Conference on Sensing, Signal Processing and Security (ICSSS).

Sazonov, E. S., Fulk, G., Hill, J., Schutz, Y., & Browning, R. (2011). Monitoring of posture
allocations and activities by a shoe-based wearable sensor. IEEE Transactions on Biomedical
Engineering, 58(4), 983–990. doi:10.1109/TBME.2010.2046738 PMID:20403783

Seng & Mirisaee. (2009). Evaluation of a content-based retrieval system for blood cell images
with automated methods. Journal of Medical Systems, 35, 571–578.

235
Compilation of References

Seo, H., & Khuzani, M. B. (2020). Machine learning techniques for biomedical image segmentation:
An overview of technical aspects and introduction to state-of-art applications. American Association
of Physicists in Medicine, 45(5), 148–167. PMID:32418337

Seyeddain, O., Kraker, H., Redlberger, A., Dexl, A. K., Grabner, G., & Emesz, M. (2014). Reliability
of automatic biometric iris recognition after phacoemulsification or drug-induced pupil dilation.
European Journal of Ophthalmology, 24(1), 58–62. doi:10.5301/ejo.5000343 PMID:23873488

Shallu, R. M., & Mehra, R. (2018). Breast cancer histology images classification:Training from
scratch or transfer learning? ICT Express, 4(4), 247–254. doi:10.1016/j.icte.2018.10.007

Shelke, M. P. K., Sontakke, M. S., & Gawande, A. (2012). Intrusion detection system for cloud
computing. International Journal of Scientific & Technology Research, 1, 67–71.

Singh, S., Saini, S., & Singh, M. (2012). Cancer Detection using Adaptive Neural Network.
International Journal of Advancements in Research and Technology, 1(4).

Smart Sensors Ltd. (2013). MIRLIN SDK, 2, 23.

Sorensen, Shaker, & Bruijne. (2010). Quantitative analysis of pulmonary emphysema using local
binary patterns. IEEE Trans. Med. Imaging, 29(2), 559-569.

Suckling, J., Parker, J., Dance, D., Astley, S., Hutt, I., Boggis, C., Ricketts, I., Stamatakis, E.,
Cerneaz, N., Kok, S., Taylor, P., Betal, D., & Savage, J. (1994). The mammographic image analysis
society digital mammogram database. In 2nd International Workshop on Digital Mammography.
Excerpta Medica.

Suganya, R., & Shanthi, R. (2012). Fuzzy C- Means Algorithm. RE:view, 2(11), 1–3.

Sung, M., Marci, C., & Pentland, A. (2005). Carl Marci, Alex Pentland, “Wearable feedback
systems for rehabilitation. Journal of Neuroengineering and Rehabilitation, 2(1), 1–12.
doi:10.1186/1743-0003-2-17

Sun, S., Li, W., & Kang, Y. (2015). Lung Nodule Detection Based on GA and SVM. 8th
International Conference on Bio Medical Engineering and Informatics (BMEI 2015).

Sutra, G., Dorizzi, B., Garcia-Salitcetti, S., & Othman, N. (2013, April 23). A biometric reference
system for iris. OSIRIS version 4.1. Available: http://svnext. it-sudparis.eu/svnview2-eph/ref
syst/Iris Osiris v4.1/

236
Compilation of References

Swapnil, R. T. (2016). Detection of brain tumor from MRI images by using segmentation &
SVM. World Conference on Futuristic Trends in Research and Innovation for Social Welfare
(Startup Conclave).

Tang, H., Wu, E. X., Ma, Q. Y., Gallagher, D., Perera, G. M., & Zhuang, T. (2000). MRI brain
image segmentation by multi-resolution edge detection and region selection. Computerized Medical
Imaging and Graphics, 24(6), 349–357. doi:10.1016/S0895-6111(00)00037-9 PMID:11008183

Thuy, Hai, & Thai. (n.d.). Image Classification using Support Vector Machine and Artificial
Neural Network. Academic Press.

Trokielewicz, M., Czajka, A., & Maciejewicz, P. (2014). Cataract influence on iris recognition
performance. Proc. SPIE 9290, Photonics Applications in Astronomy, Communications, Industry,
and High-Energy Physics Experiments. doi:10.1117/12.2076040

Trujillo, C., Piedrahita-Quintero, P., & Garcia-Sucerquia, J. (2020). Digital lensless holographic
microscopy: Numerical simulation and reconstruction with ImageJ. Applied Optics, 59(19),
5788–5795. doi:10.1364/AO.395672 PMID:32609706

Unique Identification Authority of India. (n.d.). AADHAAR. Available: https://uidai.gov.in/

what-is-aadhaar.html

Urban, G. (2014). Multi-modal brain tumor segmentation using deep convolutional neural networks.
MICCAI BraTS (Brain Tumor Segmentation) Challenge. Proceedings, 31-35.

Veras, R. (2015). SURF descriptor and pattern recognition techniques in automatic identification
of pathological retinas. 2015 Brazilian Conference on Intelligent Systems.

Vijaya, Suhasini, & Priya. (n.d.). Automatic detection of lung cancer in CT images. IJRET:
International Journal of Research in Engineering and Technology.

Vijay, J., & Subhashini, J. (2013). An Efficient Brain Tumor Detection Methodology Using K-Means
Clustering Algorithm. International conference on Communication and Signal Processing.

Vincent, P., Larochelle, H., Bengio, Y., & Manzagol, P.-A. (2008). Extracting and Composing
Robust Features with Denoising Autoencoders. In 25th International Conference on Machine
Learning. New York: ACM. 10.1145/1390156.1390294

Wadhwani & Saraswat. (2009). Classification of breast cancer using artificial neural network.
Current Research in Engineering. Science and Technology Journals.

Wang. (2017). The multimodal brain tumor image segmentation based on convolutional neural
networks. ICCIA.

237
Compilation of References

Wang, D., Yuan, F., & Sheng, H. (2010). An Algorithm for Medical Imaging Identification based
on Edge Detection and Seed Filling. In 2010 International Conference on Computer Application
and System Modeling (ICCASM 2010). Taiyuan: IEEE.

Wang, Y., & Cao, Y. (2019). Quick leukocyte nucleus segmentation in leukocyte counting.
Computational and Mathematical Methods in Medicine, 2019, 2019. doi:10.1155/2019/3072498
PMID:31308855

Wells, W. M., Grimson, W. E. L., Kikinis, R., & Jolesz, F. A. (1996). Adaptive segmentation
of MRI data. IEEE Transactions on Medical Imaging, 15(4), 429–442. doi:10.1109/42.511747
PMID:18215925

Werlinger, R., Hawkey, K., Muldner, K., Jaferian, P., & Beznosov, K. (2008). The challenges of
using an intrusion detection system: is it worth the effort? Proceedings of the 4th symposium on
Usable privacy and security, 107-118. 10.1145/1408664.1408679

Wu, Y., Wang, N., Zhang, H., Qin, L., Yan, Z., & Wu, Y. (2010). Application of Neural Networks
in the Diagnosis of Lung Cancer by Computed Tomography. Sixth International Conference on
Natural Computation. 10.1109/ICNC.2010.5583316

Xiao, Y., Wu, J., Lin, Z., & Zhao, X. (2018). Breast Cancer Diagnosis Using an Unsupervised
Feature Extraction Algorithm Based on Deep Learning. In 2018 37th Chinese Control Conference
(CCC). IEEE. 10.23919/ChiCC.2018.8483140

Xing, T., Huang, D., Xu, L., Chung, C.-J., & Khatkar, P. (2013). Snortflow: A openflow-based
intrusion prevention system in cloud environment. Research and Educational Experiment Workshop
(GREE), 89-92. 10.1109/GREE.2013.25

Xu, Q., & Zhang, L. (2015). The Effect of Different Hidden Unit Number of Sparse Autoencoder.
In The 27th Chinese Control and Decision Conference (2015 CCDC). IEEE. 10.1109/
CCDC.2015.7162335

Xue, Long, & Antani, Jeronimo, & Thoma. (2008). A Web-accessible content-based cervicographic
image retrieval system. Proceedings of the Society for Photo-Instrumentation Engineers, 6919.

Xueming, Hua, Chen, & Liangjun. (2011). PLBP: An effective local binary patterns texture
descriptor with pyramid representation. Pattern Recognition, 44, 2502–2515.

Xu, S., Zhang, Y., Jia, L., & Kyle, E. (2014). Soft microfluidic assemblies of sensors, circuits,
and radios for the Skin. Science, 344(6179), 70–74. doi:10.1126cience.1250169 PMID:24700852

Yadav, A., & Kumar, N. (2016). A Survey of Authentication Methods in Cloud Computing.
International Journal of Innovative Research in Computer and Communication Engineering,
4, 19529–19533.

Yao, C.-H., & Chen, S.-Y. (2003). Retrieval of translated, rotated and scaled color textures. Pattern
Recognition, 36(4), 913–929. doi:10.1016/S0031-3203(02)00124-3

238
Compilation of References

Yarlagadda, V. K., & Ramanujam, S. (2011). Data security in cloud computing. Journal of
Computer and Mathematical Sciences, 2, 1–169.

Yasiran, S. S., Salleh, S., & Mahmud, R. (2016). Haralick texture and invariant moments
features for breast cancer classification. In AIP Conference Proceedings. AIP Publishing.
doi:10.1063/1.4954535

Ye, S., Zheng, S., & Hao, W. (2010). Medical image edge detection method based on adaptive
facet model. In 2010 International Conference on Computer Application and System Modeling
(ICCASM 2010). Taiyuan: IEEE.

Yezzi, A. J., Kichenassamy, S., Kumar, A., Olver, P., & Tannenbaum, A. (1997). A geometric
snake model for segmentation of medical imagery. IEEE Transactions on Medical Imaging,
16(2), 199–209. doi:10.1109/42.563665 PMID:9101329

Yi, D. (2016). 3-D convolutional neural networks for glioblastoma segmentation. arXiv preprint
arXiv:1611.04534.

Yuan, X., Zhou, H., & Shi, P. (2007). Iris recognition: A biometric method after refractive surgery.
Journal of Zhejiang University. Science A, 8(8), 1227–1231. doi:10.1631/jzus.2007.A1227

Yurdakul, Subathra, & Georgec. (2020). Detection of Parkinson’s Disease from gait using
Neighborhood Representation Local Binary Patterns. Biomedical Signal Processing and Control,
62.

Zakeri, F. S., Behnam, H., & Ahmadinejad, N. (2010). Classification of benign and malignant
breast masses based on shape and texture features in sonography images. Journal of Medical
Systems, 36(3), 1621–1627. doi:10.100710916-010-9624-7 PMID:21082222

Zhang, B., Gao, Y., Zhao, S., & Liu, J. (2010). Local derivative pattern versus local binary
pattern: Face recognition with higher-order local pattern descriptor. IEEE Transactions on Image
Processing, 19(2), 533–544.

Zhang, D., Yang, G., & Li, F. (2020). Detecting seam carved images using uniform local binary
patterns. Multimedia Tools and Applications, 79, 8415–8430. doi:10.100711042-018-6470

Zheng, S. (2015). Conditional random fields as recurrent neural networks. Proceedings of the
IEEE international conference on computer vision. 10.1109/ICCV.2015.179

Zhou, H., Stone, T., Hu, H., & Harris, N. (2008). Use of multiple wearable inertial sensors in
upper limb motion tracking. Medical Engineering & Physics, 30(1), 123–133. doi:10.1016/j.
medengphy.2006.11.010 PMID:17251049

Zikic, D. (2014). Segmentation of brain tumor tissues with convolutional neural networks.
Proceedings MICCAI-BRATS, 36-39.

Zulpe & Chowhan. (2011). Statical Approach For MRI Brain Tumor Quantification. International
Journal of Computer Applications, 35(7).

239
240

About the Contributors

Kalaivani Anbarasan, an Academician and Researcher, received her M.C.A

degree from Presidency College, Madras University, 1998. She received her M.Phil
Computer Science from Alagappa University and also received M.E Gold Medal
Computer Science and Engineering degree from St. Peter’s University. She has
completed her prestigious Ph.D degree in Computer Science Engineering from
Anna University Chennai, 2017. She has gained her 17+ years of teaching experi-
ence from removed universities and Engineering Colleges. She received South
Indian “Best Non-Circuit Faculty Award” from ASDF Organization during April
2018. She also received Best Teacher award from Lions Club International during
October 2018. She received Best Paper Award in International / National Confer-
ences. She is acting as a Trustees for P.T. Lee Chengalvaraya Naicker Trust by
Honourable High Court of Madras for the period of three years (2017-2020). She
is appointed as a Governing Council Member and School Committee Member for
P.T. Lee Chengalvaraya Naicker Trust for the period of three years (2017-2020). She
has attended national workshop, FDP and Training programs and also completed
NPTEL online certification course. She has published papers in high reputed journal
and presented papers in IEEE and Springer Conference. She acted as a resource
person and delivered lecturers to both faculties and students on research focus and
upcoming technologies. She served as a technical advisory committee member for
reputed international conferences. She is acting as an reviewer and editor for high
reputed international journals and international conferences.

***

Arthi B. holds a Ph.D. degree in the field of Computer Science and Engineering
from Anna University. She has 14 years of experience in teaching. Her area of interest
includes Software Engineering, IOT, Cloud Computing and Green Computing. She
has published several articles in various reputed journals. She has presented papers
in various national and international conferences and attended many workshops,
seminars and faculty development programs in order to be in track with the chang-
About the Contributors

ing technology and teaching methodology. She is a member of various scientific

and professional bodies. She has been awarded the IET Inspiring Young Teacher
Award for the year 2016-2017 for the IET Chennai Local Network.

D. K. Chaturvedi is working in Dept. of Elect. Engg, Faculty of Engg, D.E.I.,

Dayalbagh, Agra since 1989. Presently he is Professor. He did his B.E. from Govt.
Engineering College Ujjain, M.P. then he did his M.Tech. and Ph.D. from D.E.I.
Dayalbagh. He is gold medalist and received Young Scientists Fellowship from
DST, Government of India in 2001-2002 for post doctorial research at Univ. of
Calgary, Canada. Also, he had research collaboration with different organizations
at national and international level. He is the Fellow - The Institution of Engineers
(India), Fellow - Aeronautical Society of India, Fellow - IETE, Sr. Member IEEE,
USA and Member of many National and International professional bodies such as
IET, U.K., ISTE, Delhi, ISCE, Roorkee, IIIE, Mumbai and SSI etc. The IEE, U.K.
recognized his work in the area of Power System Stabilizer and awarded honorary
membership to him in 2006. He did many R&D projects of MHRD, UGC, AICTE
etc. and consultancy projects of DRDO. He contributed in the national mission of
ICT of Govt. of India as Virtual Power Lab Developer. He has guided 10 Ph.Ds.,
65 M.Tech. Dissertations and published more than 300 International and National
Papers. He has chaired and Co-Chaired many International and National Confer-
ences. He is referee of many International Journals including IEE Proceedings and
IEEE Transactions. He is Head of Dept. of Footwear Technology, Convener, Faculty
Training and Placement Cell, and Advisor, IEI Students’ Chapter (Elect. Engg.),
D.E.I. Dayalbagh, Agra.

Washington Wagner da Silva holds a PhD in Computer Science from the Federal
University of Pernambuco - UFPE (2017). He holds a Master’s degree in Computer
Science from the Federal University of Pernambuco - UFPE (2011). He holds a
degree in Systems Analysis from the Salgado de Oliveira University - UNIVERSO
(2004). He has a postdoctoral degree in the Department of Biomedical Engineering
of the Federal University of Pernambuco - UFPE (10/2017 until 10/2019) having as
supervisor Professor. Dr. Wellington Pinheiro dos Santos. He was a Test Engineer
of the CIn / Motorola project (from May 10, 2006 to October 31, 2007). He has
experience in Computer Science, acting on the following subjects: Software Testing
Engineering, Artificial Intelligence, Artificial Neural Networks, Hybrid Intelligent
Systems, Handwriting Character Recognition, Pattern Recognition and Biomedical
Engineering.

241
About the Contributors

Maira de Santana has a Master’s degree in Biomedical Engineering at the Fed-

eral University of Pernambuco (UFPE) and member of the Biomedical Computing
Research Group. She holds a degree in Biomedical Engineering from the Federal
University of Pernambuco (2017). She has fluency in Portuguese (native language)
and English, as well as basic knowledge of Spanish and German. She completed
an internship in Clinical Engineering at Hospital das Clínicas de Pernambuco
(09/2016 - 01/2017). She was a Scholarship for Science for Borders Program of the
Federal Government / CAPES in the United States for one year (edict 180: 08/2015 -
08/2016), of which nine months (08/2015 - 05/2016) were dedicated to She has a BA
in Biomedical Engineering at the University of Alabama at Birmingham (UAB), AL
(USA) and in the last three months (05/2016 - 08/2016) she worked as a researcher
at the Carl E Ravin Advanced Imaging Laboratories (RAILabs) - Duke University,
NC, USA - deepening specific knowledge of the area of image processing (sub-area
of Biomedical Engineering) and acquiring experience in laboratory stage, scientific
production, programming in MATLAB language and Office package.

Wellington Pinheiro dos Santos received a bachelor’s degree in Electrical

Electronics Engineering (2001) and MSc in Electrical Engineering (2003) from the
Federal University of Pernambuco, and Ph.D. in Electrical Engineering from the
Federal University of Campina Grande (2009). He is currently a Professor of the
Department of Biomedical Engineering at the Federal University of Pernambuco,
acting in Undergraduate and Graduate Programs in Biomedical Engineering. He is
a member of the Graduate Program in Computer Engineering from the Polytechnic
School of Pernambuco, University of Pernambuco, since 2009. He has experience
in the area of Computer Science, acting on the following themes: digital image pro-
cessing, pattern recognition, computer vision, evolutionary computation, numerical
methods of optimization, computational intelligence, computer graphics, virtual
reality, game design and applications of Computing and Engineering in Medicine
and Biology. He is a member of the Brazilian Society of Biomedical Engineering
(SBEB), the Brazilian Society of Computational Intelligence, and the International
Federation of Medical and Biological Engineering (IFMBE).

Balanagireddy G. is currently working as an Assistant Professor in the depart-

ment of Electronics and Communication Engineering, Rajiv Gandhi University of
Technologies-AP Dr.APJ Abdul Kalam Campus Ongole, India. He did his M.Tech
with Jawaharlal Nehru Technical University Hyderabad, India and pursuing Ph.D
with Visveswaraiah Technological University Belgaum, India. His area of interest
includes Nanoelectronics, MEMS and Ad-hoc networks. He has very good achieve-
ments in NPTEL online courses. He is an active member of ISSS, ISTE and The
Indian science congress Association.

242
About the Contributors

Padmapriya Govindhan received the M.E., degree in Computer Science and

Engineering from K. S. Rangasamy College of Technology, Tiruchengode in 2008
and Ph.D in Information and Communication Engineering at Anna University, Chen-
nai in 2016. She worked as Assistant Professor in Department of Computer Science
and Engineering at K.S.R. College of Engineering (Autonomous), Tiruchengode
from 2008, Associate Professor in Department of Computer Science and Engineer-
ing at Vidyaa Vikas College of Engineering and Technology, Tiruchengode from
2016 and currently working as Associate Professor at the Department of Computer
Science and Engineering, at Saveetha School of Engineering, SIMATS, Chennai,
India.. She has published more than 28 papers in refereed journals and conference
proceedings. Her current research interest includes Data Mining, Text Mining,
Information Retrieval, Natural Language processing Deep Learning and Neural
Networks. She is a member of ISTE, IAENG, and CSTA.

Mayank Gupta is acting as System and IT Analyst in Tata Consultancy services,

Noida and expert of Data sciences and Business Analytics. He has skill to visualize
the situations from different perspectives and explore the real facts through critical
Analysis. He has deep interest in Human health domains.

Ananthajothi K. (1983) is an Assistant Professor in the Department of Computer

Science and Engineering at Misrimal Navajee Munoth Jain Engineering College,
Chennai,(INDIA). He obtained his Master degree (M.E) in Computer Science and
Engineering and Ph.D in Computer Science and Engineering from Anna University
in the year 2010 and 2020 respectively. His research focuses are Data Mining and
Big-data. He published book title of Theory of computation.

Sujatha Kesavan is presently working as Professor in EEE, Department at Dr.

M.G.R Educational and Research Institute, Chennai, Tamil Nadu, India and heading
the Research centre ‘Center for Electronics, Automation and Industrial Research
(CEAIR). She has 20 years of teaching experience in various Engineering colleges.
She completed her BE in the year 1999 from Bharathiyar University, ME in 2004
and Ph. D in 2011 from Anna University. She has presented/published papers in
National/International conferences/journals and also published many books with
Elsevier and Springer publisher. She is also a reviewer for journals published by
Springer and Elsevier publishers. Presently doing her research in the area of Image
Processing for Process Control. She is awarded the Best Researcher award for the
academic year 2011–2012 and 2014 by IET. Also obtained travel grant from DST
in 2014 for attending the conference. She is also awarded the young researcher
award at the international conference at China in year 2015. She has also published
4 patents including one international patent with the Chinese University at Huaiyin

243
About the Contributors

Institute of Technology, China and also initiated international cooperation research

between Huaiyin Institute of Technology, China and Dr.MGR Educational and
Research Institute.

Aruna M. holds Ph.D. Degree in the Faculty of Information and Communica-

tion Engineering from Anna University, Chennai. She has 14 years of experience
in teaching and currently working as Assistant Professor (SG) in the Department of
Computer Science and Engineering at Saveetha School of Engineering, SIMATS,
Chennai, India. Her area of interest includes Cloud Computing, IoT, Artificial
Intelligence and Machine Learning, Green Computing and Software Engineering,
Engineering. She has published several articles in various reputed Journals and
Conferences proceedings in order to be in track with the changing technology and
teaching methodology. She is a member of various Scientific and Professional bod-
ies such as ISTE, IAENG, CSTA, SDIWC.

Malathi M. received her B.E degree in Electronics and Instrumentation engineer-

ing from Madurai Kamaraj University, in 2004, M.E degree in Applied Electronics
from Sathyabama University in 2011. She is currently working as Associate Pro-
fessor in the Department of Electronics and communication Engineering, Chennai
Institute of Technology, Chennai, India. She has completed Ph.D from Dr.MGR.
Educational Research Institute in 2018. She obtained Anna university guide ship
under the area of Medical Image processing.

Deepa Narayanan is currently working as Assistant Professor in Saveetha

School of Engineering in Computer Science and Engineering Department. She has
published papers in several Scopus, journals and presented papers in both national
and international conferences. Her areas of interest includes image processing, big
data and video analytics. She has guided many students in various domains and has
motivated students for paper presentation and publication.

Elakkiya R. is an Assistant Professor in the Department of Computer Science and

Engineering, School of Computing, SASTRA University, Thanjavur. She received
her Ph.D. (Information & Communication Engineering) and an M.E (Software
Engineering) from Anna University, Chennai, India in 2018 and 2012, respectively.
She received her Bachelor’s degree in Computer Science & Engineering from Anna
University, Chennai, India in 2010. She has more than 7 years of research experi-
ence in Machine Learning and Computer vision. She has published more than 20
research articles in leading journals, conference proceedings and book including
IEEE, Elsevier and Springer. Currently, she is an editor of Information Engineering
and Applied Computing journal and also, she is a Life time member of International
Association of Engineers.
244
About the Contributors

Rohit Rastogi received his B.E. degree in Computer Science and Engineer-
ing from C.C.S.Univ. Meerut in 2003, the M.E. degree in Computer Science from
NITTTR-Chandigarh (National Institute of Technical Teachers Training and Research-
affiliated to MHRD, Govt. of India), Punjab Univ. Chandigarh in 2010. Currently
he is pursuing his Ph.D. In computer science from Dayalbagh Educational Institute,
Agra under renowned professor of Electrical Engineering Dr. D.K. Chaturvedi in
area of spiritual consciousness. Dr. Santosh Satya of IIT-Delhi and dr. Navneet Arora
of IIT-Roorkee have happily consented him to co supervise. He is also working
presently with Dr. Piyush Trivedi of DSVV Hardwar, India in center of Scientific
spirituality. He is a Associate Professor of CSE Dept. in ABES Engineering. Col-
lege, Ghaziabad (U.P.-India), affiliated to Dr. A.P. J. Abdul Kalam Technical Univ.
Lucknow (earlier Uttar Pradesh Tech. University). Also, He is preparing some
interesting algorithms on Swarm Intelligence approaches like PSO, ACO and BCO
etc. Rohit Rastogi is involved actively with Vichaar Krnati Abhiyaan and strongly
believe that transformation starts within self.

Anusuya S. is Professor & Head Department of Information Technology Saveetha

School of Engineering Saveetha Institute of Medical and Technical Sciences.

Kannan S. is Asst. Professor, Dept of Computer Science and Engineering Saveetha

School of Engineering, Saveetha Institute of Medical and Technical Sciences.

Ganesh T. R. is currently working as a professor, Department of Electronics

and Communication Engineering, Muthayammal Engineering College, Rasipuram,
Namakkal District. He has obtained his A M I E degree in Electronics and Com-
munication Engineering from The Institution of Engineers (India) in 1993. He re-
ceived his ME degree in Applied Electronics from Karunya Institute of Technology,
Coimbatore, in 2001. He obtained his Ph.D from Anna University, Chennai, India
in 2014 in the field of Medical Image Processing. He has authored over Seventy
eight research publications in international and national journals conferences and
books. Published 1 Patent and filed 1 patent. His special areas of interest are Signal
Processing, Wireless Networks, Image Processing, Control system and Biomedi-
cal Instrumentation. He guided three Ph.D Scholars and pursuing eight scholars
from Anna University, Chennai, India. Reviewed around ten referred journals like
Springer, Elseware, Biomedical research, and Inder Science publishers. He received
distinguish faculty award in Venus international Foundation, Chennai, India .He is
a editorial board member of internal scientific journal of Contemporary research in
engineering science and Management .He is a Life time member of IETE and IE.

245
About the Contributors

Devi Thiyagarajan is currently working as Assistant Professor in Saveetha

School of Engineering in Computer Science and Engineering Department. She
has published papers in several Scopus, Web of Science, SCI indexed journals and
presented papers in both national and international conferences. Her areas of interest
includes image processing, cloud computing and video analytics. She has guided
many students in various domains and has motivated students for paper presenta-
tion and publication.

Sudha V. is currently pursuing her Ph.D in Bio-medical Image Processing under

Anna University, Chennai registered in 2017. She has completed her P.G in Anna
University of Technology, Coimbatore in 2012. Currently, She is working as As-
sistant Professor in the Department of Electronics & Communication Engineering at
Sona College of Technology, (Autonomous) Salem. She possess 9 years of teaching
experience and has made significant contributions in the area of Deep learning using
Convolutional neural networks, Medical image analysis for Diabetic Retinopathy.
She has made significant publications in International and National journals.

246
247

Index

3D reconstruction 155 35, 64, 91-93, 95, 99-100, 108, 112,

188, 194, 197, 202, 205
A Fuzzy C Means 31, 68, 70-71, 205

active contour segmentation 194, 196-198, H

201, 203, 205
healthcare 7, 35, 37-41, 47-48, 50, 56, 58-
B 60, 135, 176, 212-213, 217-218
HMM 37, 48
Blind De-convolution 1-2, 10-11
I
C
image processing 2, 8, 11-12, 33, 35, 64,
classification 6-8, 33, 35, 43-46, 48, 50, 77, 80, 92-93, 100, 118, 136, 138, 153,
52-55, 58, 64, 67, 72-74, 76, 78-80, 82- 156, 164, 191, 201
92, 94, 96-97, 99-101, 104, 107-109, intrusion detection system 180, 191-193
112-114, 117, 119, 129-132, 134-136,
138-139, 153, 156, 160, 162, 213-214 K
cloud computing 164-169, 172-173, 175-
178, 180-181, 183, 191-193 K means clustering 194, 196-197, 205
clustering techniques 61, 194, 205 Key words: Image Processing 164
colocalization 137-141, 144-145, 152-153 K-Means 7, 61, 65, 67, 69-71, 73-74, 95,
Content Based Image Retrieval 114, 116 97, 99, 120, 156-157, 163, 194, 196,
Convolutional Neural Network (CNN) 198-199, 205-206
61, 65 KNN 37, 52, 55-56
correlation coefficient 33, 137-141, 143-
146, 148-151 L
E leukocyte segmentation 137
local binary pattern 114, 116-117, 120,
eye disease 1-2, 10, 12, 31-34 135-136

F M
feature extraction 2, 7, 9-11, 15, 27-30, 33, machine learning 6, 37-45, 48, 56-60, 79,
Index

91, 96, 107, 132, 197-198, 206, 217 100, 103-104, 112-113, 135, 138-139,
Magnetic Resonance Image (MRI) 61 153-157, 159-160, 163, 194, 196-198,
Magnetic Resonance Imaging 102, 127, 201, 203, 205-207
155-156, 194-195, 210 Segmentation FCM clustering 92
medical images 12, 104, 114-117, 119-120, spine tumour 194-195, 197-198, 205
127, 198 Support Vector Machines 46, 164, 185, 214
Medical resonance imaging 155 SVM classifier 53, 67, 74, 84, 93-94, 155,
morphological operations 139, 156, 159 158-161, 205
SVM,PCA 37
P
T
Principal Component Analysis (PCA) 2,
10, 14, 48 tuberculosis 92-94, 97, 100
tumor 61-65, 67-75, 87, 94, 100-102, 113,
R 155-156, 159-163, 205-206, 209, 215,
217-218
Random Forest 6, 9, 83-84, 92-93, 96-97,
197 W
S wavelet transform 1, 7, 9-12, 27-30, 33, 78,
80, 82, 88, 119, 157
Secure Cloud Model 164, 191
segmentation 6-9, 34-36, 61, 64-65, 67,
70-75, 79, 86-87, 89, 92, 95, 97, 99-

248

Cândida Lopes Alves - Katie Moraes Almondes - Gilberto Sousa Alves - Advances in Neuroscience, Neuropsychiatry, and Neurology-IGI Global (2024)
No ratings yet
Cândida Lopes Alves - Katie Moraes Almondes - Gilberto Sousa Alves - Advances in Neuroscience, Neuropsychiatry, and Neurology-IGI Global (2024)
324 pages
Artificial-Intelligence-in-Healthcare-Part-1 3
No ratings yet
Artificial-Intelligence-in-Healthcare-Part-1 3
17 pages
Healthcare Solutions Using Machine Learning and Informatics 1st Edition Scribd PDF Download
100% (10)
Healthcare Solutions Using Machine Learning and Informatics 1st Edition Scribd PDF Download
15 pages
AI Presentation
No ratings yet
AI Presentation
16 pages
AI FOR CLINICAL INFORMATION SYSTEM - Edit
No ratings yet
AI FOR CLINICAL INFORMATION SYSTEM - Edit
98 pages
Deep Learning, Machine Learning and Iot in Biomedical and Health Informatics (Sujata Dash, Subhendu Kumar Pani Etc.) (Z-Library)
No ratings yet
Deep Learning, Machine Learning and Iot in Biomedical and Health Informatics (Sujata Dash, Subhendu Kumar Pani Etc.) (Z-Library)
382 pages
The State of Artificial Intelligence in Medical Imaging
No ratings yet
The State of Artificial Intelligence in Medical Imaging
44 pages
Previewpdf
No ratings yet
Previewpdf
81 pages
unit1-AI in Healthcare
No ratings yet
unit1-AI in Healthcare
37 pages
Lec 23 Advancements in Medical Imaging A Look Into Future
No ratings yet
Lec 23 Advancements in Medical Imaging A Look Into Future
37 pages
Quick Download Smart Medical Imaging For Diagnosis and Treatment Planning 1st Edition EPUB DOCX PDF Download
No ratings yet
Quick Download Smart Medical Imaging For Diagnosis and Treatment Planning 1st Edition EPUB DOCX PDF Download
17 pages
Ai in Health Care - Unit - II
No ratings yet
Ai in Health Care - Unit - II
57 pages
Artificial Intelligence/ Machine Learning in Nuclear Medicine and Hybrid Imaging
No ratings yet
Artificial Intelligence/ Machine Learning in Nuclear Medicine and Hybrid Imaging
216 pages
3-IJCI Vol. 3 No. 9-Sept 2024-Paper2-Ms. Rahaf
No ratings yet
3-IJCI Vol. 3 No. 9-Sept 2024-Paper2-Ms. Rahaf
31 pages
ArtificialIntelligence and Machine Learning Rohit Tanwar
0% (1)
ArtificialIntelligence and Machine Learning Rohit Tanwar
155 pages
Аlех Кhаng Technologies in the Medical
No ratings yet
Аlех Кhаng Technologies in the Medical
458 pages
Trends of Artificial Intelligence and Big Data For E-Health by Houneida Sakly
No ratings yet
Trends of Artificial Intelligence and Big Data For E-Health by Houneida Sakly
256 pages
Explainable AI in Health Informatics: Rajanikanth Aluvalu Mayuri Mehta Patrick Siarry
No ratings yet
Explainable AI in Health Informatics: Rajanikanth Aluvalu Mayuri Mehta Patrick Siarry
287 pages
Artificial Intelligence in Healthcare Material
No ratings yet
Artificial Intelligence in Healthcare Material
29 pages
10.1201 9780367548445 Previewpdf
No ratings yet
10.1201 9780367548445 Previewpdf
86 pages
Stephan T. Artificial Intelligence in Medicine 2025
No ratings yet
Stephan T. Artificial Intelligence in Medicine 2025
265 pages
PSG College Artificial Intelligence in Health Care Indutry Co Sai
No ratings yet
PSG College Artificial Intelligence in Health Care Indutry Co Sai
9 pages
Assistive Technology Intervention in Healthcare - 1st Edition Full Digital Edition
No ratings yet
Assistive Technology Intervention in Healthcare - 1st Edition Full Digital Edition
14 pages
Advancement, Opportunities, and Practices in Telehealth Technology Full Book Download
100% (17)
Advancement, Opportunities, and Practices in Telehealth Technology Full Book Download
14 pages
Buy Ebook New Technologies For Advancing Healthcare and Clinical Practices 1st Edition Joseph Tan Cheap Price
No ratings yet
Buy Ebook New Technologies For Advancing Healthcare and Clinical Practices 1st Edition Joseph Tan Cheap Price
67 pages
Smart Medical Diagnosis System
100% (1)
Smart Medical Diagnosis System
34 pages
Chapter Report
No ratings yet
Chapter Report
44 pages
Machine Learning Application in Healthcare
No ratings yet
Machine Learning Application in Healthcare
12 pages
B9780443247880000078
No ratings yet
B9780443247880000078
1 page
Assistive Technology Intervention in Healthcare - 1st Edition One-Click Download
100% (13)
Assistive Technology Intervention in Healthcare - 1st Edition One-Click Download
14 pages
AI and ML in Medicine Research Paper
No ratings yet
AI and ML in Medicine Research Paper
2 pages
Lecture Four
No ratings yet
Lecture Four
28 pages
AI and Medical Imaging Technology Evolution Impacts
No ratings yet
AI and Medical Imaging Technology Evolution Impacts
13 pages
New Technologies For Advancing Healthcare and Clinical Practices PDF
100% (18)
New Technologies For Advancing Healthcare and Clinical Practices PDF
14 pages
Applications of Artificial Intelligence in Medicine
No ratings yet
Applications of Artificial Intelligence in Medicine
9 pages
Access Driving Smart Medical Diagnosis Through AI Powered Technologies and Applications Complete Digital Book
No ratings yet
Access Driving Smart Medical Diagnosis Through AI Powered Technologies and Applications Complete Digital Book
15 pages
AI Healthcare Paper With Images
No ratings yet
AI Healthcare Paper With Images
8 pages
AI in Healthcare Research Paper
No ratings yet
AI in Healthcare Research Paper
2 pages
ICT202B Paritosh
No ratings yet
ICT202B Paritosh
18 pages
Maedica 17 420
No ratings yet
Maedica 17 420
7 pages
Review Paper
No ratings yet
Review Paper
6 pages
Artificial Intelligence and Machine Learning For Medical Imaging
No ratings yet
Artificial Intelligence and Machine Learning For Medical Imaging
15 pages
Future of AI in Biomedicine and Biotechnology Scribd PDF Download
No ratings yet
Future of AI in Biomedicine and Biotechnology Scribd PDF Download
14 pages
Is Artificial Intelligence (AI) Effective in Diagnosing Diseases
No ratings yet
Is Artificial Intelligence (AI) Effective in Diagnosing Diseases
3 pages
AI in Healthcare
No ratings yet
AI in Healthcare
42 pages
Future of AI in Biomedicine and Biotechnology IGI Global, 2024
No ratings yet
Future of AI in Biomedicine and Biotechnology IGI Global, 2024
376 pages
SER Plagiarism Report
No ratings yet
SER Plagiarism Report
2 pages
The Internet of Medical Things (Iomt) and Telemedicine Frameworks and Applications
No ratings yet
The Internet of Medical Things (Iomt) and Telemedicine Frameworks and Applications
358 pages
Flyer 1
No ratings yet
Flyer 1
3 pages
Artificial Intelligence in Medicine
No ratings yet
Artificial Intelligence in Medicine
18 pages
My Final Report
No ratings yet
My Final Report
17 pages
Challenges and Opportunities in Integrating Machine Learning With Medical Imaging A Comprehensive Review
No ratings yet
Challenges and Opportunities in Integrating Machine Learning With Medical Imaging A Comprehensive Review
6 pages
Android Practical File (IT-602)
50% (2)
Android Practical File (IT-602)
59 pages
Project One
No ratings yet
Project One
1 page
192a Lt-2 LM 500 Elite Fanuc Oi TF Impm
100% (2)
192a Lt-2 LM 500 Elite Fanuc Oi TF Impm
138 pages
AI in Diagnostics, Revised
No ratings yet
AI in Diagnostics, Revised
5 pages
J of Medical Radiation Sci - 2019 - Lewis - Artificial Intelligence in Medical Imaging Practice Looking To The Future
No ratings yet
J of Medical Radiation Sci - 2019 - Lewis - Artificial Intelligence in Medical Imaging Practice Looking To The Future
4 pages
Diagnostics 13 00688
No ratings yet
Diagnostics 13 00688
3 pages
The Role of Artificial Intelligence in Radiology
No ratings yet
The Role of Artificial Intelligence in Radiology
4 pages
The Role of Artificial Intelligence in Healthcare
No ratings yet
The Role of Artificial Intelligence in Healthcare
6 pages
Report - Artificial Intelligence Applications in Medicine
No ratings yet
Report - Artificial Intelligence Applications in Medicine
12 pages
CA - Oresund Bridge
No ratings yet
CA - Oresund Bridge
29 pages
Unit V File Processing: Text Files
No ratings yet
Unit V File Processing: Text Files
26 pages
Name: Alex Tito R. Corporal Year/Course: Btvted Elx 2A
No ratings yet
Name: Alex Tito R. Corporal Year/Course: Btvted Elx 2A
3 pages
07108001
No ratings yet
07108001
96 pages
Annotated Bibliography
No ratings yet
Annotated Bibliography
18 pages
Daikool VRF Catalog.
No ratings yet
Daikool VRF Catalog.
38 pages
The Actual and Expected Reading Comprehension Ability of Grade Nine Students:Entoto Secondary School in Focus
No ratings yet
The Actual and Expected Reading Comprehension Ability of Grade Nine Students:Entoto Secondary School in Focus
143 pages
Fill Data: Qno O1 O2 O3 O4 Ans1 Ans2 Ans3 Ans4
No ratings yet
Fill Data: Qno O1 O2 O3 O4 Ans1 Ans2 Ans3 Ans4
10 pages
Fiitjee: LPP - Gravitation
100% (1)
Fiitjee: LPP - Gravitation
5 pages
(English) Фильм 14+ «История первой любви» Смотреть в HD (DownSub.com)
No ratings yet
(English) Фильм 14+ «История первой любви» Смотреть в HD (DownSub.com)
21 pages
Litere. Teme Licenta Engleza
No ratings yet
Litere. Teme Licenta Engleza
4 pages
Practical Research 1 SLHT Week 3
No ratings yet
Practical Research 1 SLHT Week 3
8 pages
Compare 2 Architects
No ratings yet
Compare 2 Architects
2 pages
Jrs Resume Revised
No ratings yet
Jrs Resume Revised
2 pages
Hyundai Motors
No ratings yet
Hyundai Motors
9 pages
Japanese Philosopher
No ratings yet
Japanese Philosopher
4 pages
Int'l Application Guidelines
No ratings yet
Int'l Application Guidelines
18 pages
Maps The Crusher
No ratings yet
Maps The Crusher
32 pages
Cie 1
No ratings yet
Cie 1
12 pages
C.14 Queens Park Urban Conservation Area
No ratings yet
C.14 Queens Park Urban Conservation Area
13 pages
Paper 81
No ratings yet
Paper 81
6 pages
Make in India Initiative
No ratings yet
Make in India Initiative
3 pages
The European Legacy: Toward New Paradigms
No ratings yet
The European Legacy: Toward New Paradigms
11 pages
Artikel Ilmiah
No ratings yet
Artikel Ilmiah
20 pages
Pikovskaya Medio
No ratings yet
Pikovskaya Medio
2 pages
Poetry
No ratings yet
Poetry
3 pages
Executive Summary: Slavica Singer and Suncica Oberman Peterka
No ratings yet
Executive Summary: Slavica Singer and Suncica Oberman Peterka
1 page
Analysis of A Poison Tree
No ratings yet
Analysis of A Poison Tree
2 pages
Cutting-Edge AI and ML Technological Solutions: Healthcare Industry
From Everand
Cutting-Edge AI and ML Technological Solutions: Healthcare Industry
Zemelak Goraga
No ratings yet

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.