Machine Learning

for Healthcare
 Machine Learning
for Healthcare
Handling and Managing Data

Edited by
Rashmi Agrawal
Jyotir Moy Chatterjee
Abhishek Kumar
Pramod Singh Rathore
Dac-Nhuong Le
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks
does not warrant the accuracy of the text or exercises in this book. This book’s use or discussion
of MATLAB® software or related products does not constitute endorsement or sponsorship by The
MathWorks of a particular pedagogical approach or particular use of the MATLAB® software.

First edition published 2021

by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

© 2021 Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, LLC

Reasonable efforts have been made to publish reliable data and information, but the author and pub-
lisher cannot assume responsibility for the validity of all materials or the consequences of their use.
The authors and publishers have attempted to trace the copyright holders of all material reproduced
in this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so
we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, access www.copyright.
com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. For works that are not available on CCC please contact mpkbookspermissions@tandf.
co.uk

Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.

ISBN: [978-0-367-35233-2] (hbk)

ISBN: [978-0-429-33013-1] (ebk)

Typeset in Palatino
by Deanta Global Publishing Services, Chennai, India
Contents

Preface..................................................................................................................... vii
Acknowledgments..................................................................................................xi
Editors.................................................................................................................... xiii
List of Contributors............................................................................................. xvii

1. Fundamentals of Machine Learning...........................................................1

Rashmi Agrawal

2. Medical Information Systems..................................................................... 17

Uday Sah, Abhushan Chataut, and Jyotir Moy Chatterjee

3. The Role of Metaheuristic Algorithms in Healthcare........................... 25

G. Uma Maheswari, R. Sujatha, V. Mareeswari, and E. P. Ephzibah

4. Decision Support System to Improve Patient Care................................ 41

V. Diviya Prabha and R. Rathipriya

5. Effects of Cell Phone Usage on Human Health and Specifically

on the Brain..................................................................................................... 53
Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi, Mehmood Naqvi
and Shakeel Ahmed

6. Feature Extraction and Bio Signals............................................................ 69

A. Mary Judith, S Baghavathi Priya, N. Kanya, and Jyotir Moy Chatterjee

7. Comparison Analysis of Multidimensional Segmentation Using

Medical Health-Care Information............................................................. 81
Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi, Memood Naqvi,
and Azeem Khan

8. Deep Convolutional Network Based Approach for Detection

of Liver Cancer and Predictive Analytics on Cloud............................... 95
Pramod H. B. and Goutham M.

9. Performance Analysis of Machine Learning Algorithm for

Healthcare Tools with High Dimension Segmentation...................... 115
Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi, Memood Naqvi
and Mamoona Humayun

v
vi Contents

10. Patient Report Analysis for Identification and

Diagnosis of Disease.................................................................................. 129
Muralidharan C., Mohamed Sirajudeen Y., and Anitha R.

11. Statistical Analysis of the Pre- and Post-Surgery in the

Healthcare Sector Using High Dimension Segmentation.................. 159
Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi, Memood Naqvi,
and Mamoona Humayun

12. Machine Learning in Diagnosis of Children with Disorders........... 175

Lokesh Kumar Saxena and Manishikha Saxena

13. Forecasting Dengue Incidence Rate in Tamil Nadu Using

ARIMA Time Series Model...................................................................... 187
S. Dhamodharavadhani, R. Rathipriya

Index...................................................................................................................... 203
Preface

The key objective of publishing this book is to explore the concepts of

machine learning, along with recent research developments in healthcare
sectors. The book includes chapters by authors who are experts in this
area and have experience of the recent research developments in the field
of healthcare using machine learning technology. It will provide in-depth
information about the applications and utilizations of machine learning in
healthcare. It will provide a better understanding of the processing of big
data from the healthcare sector as used in machine learning processes, and
it will highlight the links between machine learning and computer science
in healthcare applications from many perspectives.
Machine learning (ML) is currently causing quite a buzz, and is having a
huge impact on healthcare. Payers, providers, and pharmaceutical compa-
nies are today seeing applicability in their spaces and are taking advantage
of ML. A machine learning model is created by feeding data into a learning
algorithm. The algorithm is the place where the magic happens. There are
algorithms to determine a patient’s length of stay based on diagnosis, for
example, and that algorithm all began when someone decided to write it and
train it with true and reliable data. Over time, the model can be re-trained
with newer data, increasing the model’s effectiveness. Machine learning is
defined as when a computer has been taught to recognize patterns by pro-
viding it with data and an algorithm to help understand that data. We call
the process of learning “training” and the output that this process produces
is called a “model”. A model can be provided with new data and it can reason
with this new information based on what it has previously learned.
Machine learning models determine a set of rules using vast amounts of
computing power that a human brain would be incapable of processing. The
more data a machine learning model is fed, the more complex the rules – and
the more accurate the predictions. Whereas a statistical model is likely to
have an inherent logic that can be understood by most people, the rules cre-
ated by machine learning are often beyond human comprehension because
our brains are incapable of digesting and analyzing enormous datasets.
Deep learning is another buzzword we often hear a lot about but it is often
misunderstood. In reality, it is just a special case of machine learning algo-
rithm through artificial neural networks. A neural network is an algorithm
that was inspired by the ways a brain works and it involves many nodes (or
“neurons”) that are often connected together in layers to form a network. A
neural network must have at least two layers – a layer of inputs and a layer
of outputs. There may be many “hidden” layers between the input layer and
output layer, and these are used to extract more information by exploiting
structure in the data. A network is considered “deep” if it has more than

vii
viii Preface

one hidden layer (see the diagram opposite which illustrates the complexity
of a neural network). Neural networks are great at solving problems where
the data is highly structured – like an image of a brain scan – but are also
“black box” algorithms. In other words, it is hard for humans to comprehend
the patterns they find. Despite being around for over 50 years, neural net-
works have only become popular and feasible in the last ten years thanks to
advances in both algorithm design and computing power. While the health-
care sector is being transformed by the ability to record massive amounts of
information about individual patients, the enormous volume of data being
collected is impossible for human beings to analyze. Machine learning pro-
vides a way to automatically find patterns and reasons behind data, which
enables healthcare professionals to move towards more personalized care –
known as precision medicine. There are many possibilities for how machine
learning can be used in healthcare settings, and all of them depend on having
sufficient data and permissions to use it. Previously, alerts and recommenda-
tions for medical practice have been developed based on external studies,
and hard coded into their software. However, that can limit the accuracy
of the data because they might be from different populations and environ-
ments. Machine learning, on the other hand, can be refined using data that
is available in that particular environment. For example, anonymized patient
record information from a hospital and the area in which it serves.
One further example of how healthcare providers can take advantage of
machine learning is utilizing it in order to predict the likelihood of hospi-
tal re-admission for chronically ill patients. Identifying the patients who are
most at risk of being re-admitted means they can be given greater support
after they have been discharged. By lowering the rate of re-admission, not
only are the lives of those most at risk improved, but it also helps save pre-
cious healthcare funds, which can then perhaps be used for wellness and
prevention programs instead.
The main research issues and important concepts relating to machine learn-
ing technology in the healthcare industry are covered in more detail through-
out this book. It will provide in-depth information about handling and
managing healthcare data with machine learning methods. Throughout this
book we will try our best to show readers about the links between machine
learning and computer science applications in healthcare sector more broadly

What You’ll Learn

1. A deeper understanding of various machine learning uses and their
implementation within wider healthcare.
2. The ability to implement machine learning systems, such as cancer
detection, and enhanced deep learning.
Preface ix

3. How to select learning methods and tuning for use in healthcare.

4. How to recognize and prepare for the future of machine learning
in healthcare through best practices, feedback loops, and intelligent
agents.

Who This Book Is For

Machine intelligence for the healthcare industry is the need of the hour. It
requires adopting both new and better ways to the healthcare providers who
can take advantage of these techniques in order to improve the health of
human beings. The machine learning techniques enable healthcare provid-
ers to garner and operationalize new insights efficiently as a by-product of
an organization’s day-to-day operations. This is becoming vital to hospitals
and health systems’ ability to persist and flourish. This book will provide
in-depth information about handling and managing healthcare data with
machine learning methods. It will discuss the long-standing challenges in
healthcare informatics and provide rational ideas of how best to navigate
them.

Rashmi Agrawal (Haryana, India)

Jyotir Moy Chatterjee (Kathmandu, Nepal)
Abhishek Kumar (Himachal Pradesh, India)
Pramod Singh Rathore (Rajasthan, India)
Dac-Nhuong Le (Haiphong, Vietnam)

MATLAB® is a registered trademark of The MathWorks, Inc. For product

information, please contact:

The MathWorks, Inc.

3 Apple Hill Drive
Natick, MA 01760-2098 USA
Tel: 508 647 7000
Fax: 508-647-7001
E-mail: info@mathworks.com
Web: www.mathworks.com
Acknowledgments

I would like to acknowledge the most important people in my life, my

uncle Mr. Moni Moy Chatterjee, my father Aloke Moy Chatterjee, and my
late mother Nomita Chatterjee. This book has been my long-cherished
dream – one which could not have been turned into a reality without the
support and love of these amazing people. They have continuously encour-
aged me despite my failing to give them the proper time and attention. I am
also grateful to my friends, who have encouraged and blessed this work with
their unconditional love and patience.

Jyotir Moy Chatterjee

Department of IT
Lord Buddha Education Foundation
Kathmandu, Nepal

xi
Editors

Rashmi Agrawal is a Professor in the Department of

Computer Applications in MRIIRS, Faridabad. Dr.
Agrawal has a rich teaching experience of more than 17
years. She is UGC-NET(CS) qualified, and has com-
pleted a PhD, M. Phil, M. Tech, MSc, and MBA(IT). Her
PhD focused on the area of machine learning, and her
areas of expertise include Artificial Intelligence,
Machine Learning, Data Mining, and Operating
Systems. She has published more than 30 research
papers in various national and international conferences and journals, and
has authored many published books and chapters. She has organized vari-
ous faculty development programs and has also directly participated in
workshops and faculty development programmes. She is actively involved
in research activities, and is a lifetime member of Computer Society of India.
She has been a member of the technical programme committee of various
reputable conferences.

Jyotir Moy Chatterjee is an Assistant Professor at the

IT Department of Lord Buddha Education Foundation
(Asia Pacific University of Technology and Innovation),
Kathmandu, Nepal. Prior to this he has worked as an
Assistant Professor at the CSE Department of GD Rungta
College of Engineering and Technology (CSVTU),
Bhilai, India. He has completed an M. Tech in Computer
Science and Engineering from Kalinga Institute of
Industrial Technology, Bhubaneswar, Odisha and a B.
Tech in Computer Science and Engineering from Dr.
MGR Educational and Research Institute, Chennai. He has published 40
international research papers, two international conference papers, authored
four books, edited eight books, written 11 book chapters and has one pat-
ent to his account. His research interests include cloud computing, big data,
privacy preservation, data mining, the Internet of Things, machine learning,
and blockchain technology. He is a member of various professional societies
and international conferences.

xiii
xiv Editors

Abhishek Kumar has a PhD in computer science from

University of Madras and an M. Tech in computer sci-
ence and engineering from Government Engineering
College Ajmer at Rajasthan Technical University, Kota,
India. He has over eight years of experience in academic
teaching and has been published more than 55 times in
reputed, peer-reviewed national and international jour-
nals, books and conferences (such as by Wiley, Taylor &
Francis, Springer, Elsevier, Science Direct, Inderscience, Annals of Computer
Science, Poland, and IEEE). His research areas include: Artificial Intelligence,
image processing, computer vision, data mining, and machine learning. He
has also been on the international conference committees of many interna-
tional conferences, and is currently serving as a reviewer for IEEE and
Inderscience journals. He has authored six internationally published books
and has edited 11 books with Wiley, IGI GLOBAL, Springer, Apple Academic
Press, CRC, and more. He is also member of various national and interna-
tional professional societies in the field of engineering and research includ-
ing being a member of IEEE, ISOC (Internet Society)); IAIP (International
Association of Innovation Professionals), ICSES (International Computer
Science and Engineering Society) IAENG (International Association of
Engineers); an associate member of IRED (Institute of Research Engineers
and Doctors; a life member of ISRD (International Society for research &
Development); and an editorial board member of IOSRD. He has received the
national Sir CV Raman lifetime achievement award in 2018 in the young
researcher and faculty category.

Pramod Singh Rathore is currently pursuing his PhD

in computer science and engineering at Bundelkhand
University and is conducting ongoing research on net-
working. He has an M. Tech in computer science and
engineering from Government Engineering College
Ajmer, at Rajasthan Technical University, Kota, India.
He has been working as an Assistant Professor of the
Computer Science and Engineering Department at
Aryabhatt Engineering College and Research Centre,
Ajmer, Rajasthan and is also a visiting faculty at Government University
MDS Ajmer. He has over eight years of experience in academic teaching
and has been published more than 45 times in reputed, peer-reviewed
national and international journals, books and conferences (such as Wiley,
IGI GLOBAL, Taylor & Francis, Springer, Elsevier, Science Direct, Annals of
Computer Science, Poland, and IEEE). He has co-authored and edited many
books with many reputed publishers like Wiley, and CRC Press, USA. His
research areas include: NS2, computer networks, mining, and DBMS.
Editors xv

Dac-Nhuong Le has a PhD and is Deputy-Head of

the Faculty of Information Technology at Haiphong
University, Vietnam and Vice-Director of Information
Technology at the Apply Center of the same univer-
sity. He is a research scientist at the Research and
Development Center of Visualization & Simulation in
(CSV), Duy Tan University, Danang, Vietnam. He has
more than 45 publications in the reputed international
conferences, journals, and online book chapter contribu-
tions (indexed by: SCI, SCIE, SSCI, Scopus, ACM, and
DBLP). His areas of research include: evaluation computing and approxi-
mate algorithms, network communication, security and vulnerability,
network performance analysis and simulation, cloud computing, and bio-
medical image processing. His core work is in network security, wireless,
soft computing, mobile computing and biomedical technology. Recently, he
has been on a technical program committee, a technical reviewer, and the
track chair for international conferences such as: FICTA 2014, CSI 2014, IC4SD
2015, ICICT 2015, INDIA 2015, IC3T 2015, INDIA 2016, FICTA 2016, IC3T 2016,
ICDECT 2016, IUKM 2016, INDIA 2017, FICTA 2017, CISC 2017, ICICC 2018,
ICCUT 2018 under the Springer-ASIC/LNAI/CISC Series. Presently, he is
serving on the editorial board of international journals and he has authored
six computer science books (published by Springer, Wiley, CRC Press,
Lambert Publication, VSRD Academic Publishing, and Scholar Press).
List of Contributors

Mary Judith A. S. Dhamodharavadhani

Assistant Professor (CSE) Department of Computer Science
Loyola-ICAM College of Periyar University
Engineering & Technology Salem, India
Chennai, India
E. P. Ephzibah
Afnizanfaizal Abdullah School of Information Technology
Department of Software and Engineering
Engineering Vellore Institute of Technology
UniversitiTeknologi Malaysia-UTM Vellore, India
Johor Bharu, Malaysia
Goutham M.
Rashmi Agrawal Department of Computer Science
Manav Rachna International and Engineering
Institute of Research and Studies NIE Institute of Technology
Faridabad, India Mysuru, India

Shakeel Ahmed Mamoona Humayun

College of Computer Sciences Department of Information Systems
and IT Jouf University
King Faisal University Al-Jouf, Saudi Arabia
Hofuf, Saudi Arabia
N. Z. Jhanjhi,
Anitha R. School of Computer Science and
Department of Computer Science Engineering (SCE)
and Engineering Taylor’s University
Sri Venkateswara College of Selangor, Malaysia
Engineering
Sriperumbudur, India Mary Judith A.
Loyola-ICAM College of
Abhushan Chataut Engineering & Technology
Lord Buddha Education Foundation Chennai, India
Kathmandu, Nepal
N. Kanya
Jyotir Moy Chatterjee Dr. M.G.R Educational and Research
Lord Buddha Education Foundation Institute
Kathmandu, Nepal Chennai, India

xvii
xviii List of Contributors

Azeem Khan R. Rathipriya

American Degree Program ADP Department of Computer Science
Taylor’s University Periyar University Salem
Selangor, Malaysia Salem, India

G. Uma Maheswari Soobia Saeed

School of Information Technology Department of Software
and Engineering Engineering
Vellore Institute of Technology UniversitiTeknologi Malaysia-UTM
Vellore, India Johor Bharu, Malaysia

V. Mareeswari Uday Sah

School of Information Technology Lord Buddha Education Foundation
and Engineering Kathmandu, Nepal
Vellore Institute of Technology
Vellore, India Lokesh Kumar Saxena
Department of Mechanical
Muralidharan C. Engineering
Department of Computer Science Faculty of Engineering and
and Engineering Technology
Sri Venkateswara College of Jamia Millia Islamia
Engineering New Delhi, India
Sriperumbudur, India
Manishikha Saxena
Mehmood Naqvi Department of Education
Faculty of Engineering Technology Government of Delhi
Mohawk College Delhi, India
Hamilton, Canada
Mohamed Sirajudeen Y.
V. Diviya Prabha Department of Computer Science
Department of Computer Science and Engineering
Periyar University Salem Sri Venkateswara College of
Salem, India Engineering
Sriperumbudur, India
Pramod H. B.
Department of Computer Science R. Sujatha
and Engineering School of Information Technology
Rajeev Institute of Technology and Engineering
Hassan, India Vellore Institute of Technology
Vellore, India
S. Baghavathi Priya
Rajalakshmi Engineering College
Chennai, India
1
Fundamentals of Machine Learning

Rashmi Agrawal

CONTENTS
1.1 Introduction..................................................................................................... 1
1.2 Data in Machine Learning.............................................................................2
1.3 The Relationship between Data Mining, Machine Learning, and
Artificial Intelligence...................................................................................... 4
1.4 Applications of Machine Learning.............................................................. 5
1.4.1 Machine Learning: The Expected....................................................5
1.4.2 Machine Learning: The Unexpected...............................................5
1.5 Types of Machine Learning........................................................................... 6
1.5.1 Supervised Learning.......................................................................... 6
1.5.1.1 Supervised Learning Use Cases........................................ 9
1.5.2 Unsupervised Learning..................................................................... 9
1.5.2.1 Types of Unsupervised Learning.................................... 10
1.5.2.2 Clustering............................................................................ 11
1.5.2.3 Association Rule................................................................. 11
1.5.2.4 Unsupervised Learning Use Case................................... 13
1.5.3 Reinforcement Learning (RL) ........................................................ 13
1.6 Conclusion..................................................................................................... 14
References................................................................................................................ 15

1.1 Introduction
Machine learning is a discipline in which algorithms are applied to help
mine knowledge out of large pools of existing information. It is the sci-
ence that gives power to computers to perform without being openly pro-
grammed. “It is defined by the ability to choose effective features for pattern
recognition, classification, and prediction based on the models derived from
existing data” (Tarca and Carey 2007). According to Arthur L Samuel (1959),
“machine learning is the ability of computers to learn to function in ways
that they were not specifically programmed to do”. Many factors have con-
tributed to making machine learning a reality. These include sources of data
that are generating vast information, improved computational control for

1
2 Machine Learning for Healthcare

processing large amounts of information in fractions of time, and algorithms

which are now more reliable and efficient.
Machine learning is one of the most exciting technologies one could come
by. As is apparent from its name, machine learning offers the computer the
ability to learn, meaning it can become more like a human. It is being vigor-
ously used today, perhaps in many more ways than one would expect.

1.2 Data in Machine Learning

The data required for analysis is gathered from various sources such as web
pages, emails, IoT sensors, text files, etc. This data serves as the input needed
for machine learning algorithms to generate insights. Without data, we can’t
train any models and all contemporary research and automation would be
ineffective. Large initiatives are spending masses of money just to collect as
much specific data as possible. Data uncertainty is common in real-world
applications. Various factors like physical data generation and collection pro-
cesses, unreliable data transmission, transmission bandwidth, measurement
errors, and decision errors contribute to the uncertainty in data. This may
apply for both numerical data and categorical data (Agrawal and Ram 2015).
After collecting data, it is preprocessed and used for extracting informa-
tion and knowledge (Figure 1.1).
Now the question of how the data is used in machine learning arises. As
shown in Figure 1.2, the data is split into three parts – testing, training, and
validation data.

FIGURE 1.1
Data, information and knowledge.

FIGURE 1.2
Types of data.
Fundamentals of Machine Learning 3

Training data is applied to train machine learning models and, after com-
pletion of the training part, testing data is used for unprejudiced valuation
of the model. Validation data is used for frequent evaluation of the model
thereafter. Thus, the data plays an important role in the model building and
selection. Data has a lot of potential for organizations and almost all large-
and mid-level organizations are therefore continuously looking for ways to
utilize it (Agrawal 2020). Some of the important dimensions of big data are
described here:

1) Volume: the main characteristic feature or dimension of big data is

its sheer volume. The term volume refers to the amount of data an
organization, or an individual, collects and/or generates. Currently,
to qualify as big data, a minimum of 1 terabyte is the threshold for big
data which stores as much data as would fit on 1,500 CDs or 220 DVDs,
(or enough to store approximately 16 million Facebook images). The
vast amounts of data are generated every second. E-commerce, social
media, and various sensors produce high volumes of unstructured
data in the form of various audio, images, and video files. Today big
data is also generated by machines, networks, and human interaction
on systems and the volume of data to be analyzed is massive.
2) Variety: is one of the most attractive dimensions in technology, as
almost all information is digitized nowadays. Traditional data types
or structured data include information (such as date, amount, and
time) in a structured way which can easily fit neatly in a relational
database. Structured data is augmented by unstructured data.
Modern day data sources include Twitter feeds, YouTube videos,
audio files, MRI images, web pages, web logs, and anything else that
can be captured and stored and does not require any meta model for
its structure to access it later on.
Unstructured data is an essential concept in big data. To under-
stand the difference between structured and unstructured data, we
can compare these two types of data and see that a picture, a voice
recording, or a tweet are all different in their function and usage,
but both express ideas and thoughts based on human understand-
ing. One of the major goals of big data is to make sense from this
unstructured data.
3) Velocity: refers to the frequency of incoming data that is needing to
be processed. The flow of data is massive and continuous. The veloc-
ity of data increases over time. In 2016, approximately 5.5 million
new devices were connected every day for collecting and sharing
data. The improved data streaming potential of linked devices will
continue to gather speed in the future. Streaming applications like
Amazon Web Services Kinesis is an example of an application that
handles the velocity of data.
4 Machine Learning for Healthcare

4) Veracity: refers to the reliability of the data. Given the increasing

volume of data being generated at an unparalleled rate, it is com-
mon that the data must contain noise. Veracity refers to the trust-
worthiness of the data that needs to be analyzed. Uncertainty and
unreliability arise due to incompleteness, inaccuracy, latency, incon-
sistency, subjectivity, and deception in data.

In addition to these four dimensions, there are two additional dimensions

which are key for operationalizing data.

5) Volatility: this big data dimension refers to the length of time that
this data will remain valid and for how long it should be stored.
6) Validity: the validity dimension in big data means that data in use
should be correct and accurate. If one wants to use the results for
decision making, validity of big data sources and subsequent analy-
sis must be accurate.

1.3 The Relationship between Data Mining, Machine

Learning, and Artificial Intelligence
Figure 1.3 represents the relationship among data mining, Artificial
Intelligence (AI), data science, and machine learning. Artificial Intelligence
can be defined as the study of training computers in such a way that comput-
ers can accomplish tasks which, at present, can be done better by humans.
Machine learning is a sub-field of knowledge science that focuses on the
development of algorithms that may be learned from in order to generate
likelihoods based on the given information.

FIGURE 1.3
Relationship in AI and machine learning.
Fundamentals of Machine Learning 5

TABLE 1.1
Difference between Machine Learning and AI
Artificial Intelligence (AI) Machine Learning (ML)
Focus is given to increasing success and Main focus is to get the maximum accuracy.
not the accuracy.
The goal of AI is to imitate human In ML, the primary goal is to be trained from data
intelligence which will be used to solve on a definite task to make the most of the
complex problems. performance of the machine on this task.
AI leads to intelligence. ML leads to knowledge.
It progresses to build up an arrangement It involves developing self-learning algorithms
to imitate humans and to behave which can learn independently.
similarly in a particular circumstance.

The difference between machine learning and AI is represented by the

Table 1.1 below.

1.4 Applications of Machine Learning

Each and every time the word machine learning is used, people generally
think of “AI” and “Neural networks that can simulate human brains”, Self-
Driving Cars, and more. But machine learning is much different. Below we
expound on anticipated and unexpected aspects of contemporary comput-
ing where machine learning is enacted.

1.4.1 Machine Learning: The Expected

1. Speech Recognition
2. Computer Vision (Facial Recognition, Pattern Recognition, and
Character Recognition Techniques belong to Computer Vision)
3. Google’s Self-Driving
4. Web Search Engine
5. Photo Tagging Applications
6. Spam Detector
7. Database Mining for Growth of Automation
8. Understanding Human Learning

1.4.2 Machine Learning: The Unexpected

1. YouTube/Netflix
2. Data Mining/Big Data
6 Machine Learning for Healthcare

3. Amazon’s Product Recommendations

4. Stock Market/Housing Finance/Real Estate

1.5 Types of Machine Learning

We can define machine learning as learning from some past experiences
based on some task, and it may have one of the following types as shown in
Figure 1.4.

1.5.1 Supervised Learning
This is the most popular paradigm for machine learning, which learns from
labeled data. A function is inferred from the data that maps the input, output
pair to the target, h: f(x,y) → y, where f is the function learned from input
and output pairs x and y, respectively. It is further of two types: classifica-
tion and regression. Classification predicts categorical answers and function
acquires the class codes of different classes, that is, (0/1) or (yes/no). Naïve
Bayes, decision tree (Batra and Agrawal 2018), k nearest neighbor (Agrawal
2019), and support vector machines (SVM) are frequently used algorithms
for classification. Regression predicts the numerical response, e.g. predict-
ing the future value of stock prices. Linear Regression, neural networks, and
regularization are algorithms used for regression. Table 1.2 (A and B) shows
the difference between classification and regression.
Table 1.2A represents the classification task by showing the dataset of
a shopping store with input variables as user ID, gender, age, and salary.

FIGURE 1.4
Types of machine learning.

TABLE 1.2A
Classification
User ID Gender Age Salary Buy Product (Yes/No)
101 M 42 15k Yes
102 M 65 55k No
103 F 65 50k Yes
105 F 35 20k Yes
Fundamentals of Machine Learning 7

TABLE 1.2B
Regression
Temp Pressure Relative Humidity Wind Direction Wind Speed
17.70 988.11 39.11 192.92 2.973
24.23 988.24 19.74 318.32 0.32
22.54 989.56 22.81 44.66 0.264

Based on these input variables, the machine learning algorithm will predict
whether the customer will buy a product or not (0 for no, 1 for yes). Table 1.2B
here shows the data of a meteorological department with input variables of
temperature, pressure, relative humidity, and wind direction; after applying
regression techniques, wind speed is determined.
In classification, the goal is to predict discrete values going to a specific
class and calculate them on the basis of accuracy. This can be 0 or 1 (yes or
no) in binary classification, but in the case of multi-class classification, it is
more than one class. In regression, the output has continuous values.
Table 1.3 summarizes the supervised algorithms which are used in
machine learning.
A method of performance measurement for machine learning classifica-
tion is a confusion matrix, which has been shown in Table 1.4 for a binary
clarification problem, and Table 1.5 represents a confusion matrix for a three-
class problem. It is immensely functional for evaluating Precision, Recall,
Specificity, AUC-ROC Curve, and Accuracy.
Each entry of Tables 1.4 and1. 5 depicts the number of records from class I
predicted to be of class j. For example, C10 is the number of records from class
1 incorrectly predicted as class 0. On the other hand, C00 represents the num-
ber of records from class 0 that were correctly predicted as class 0. From the
confusion matrix we can find the total number of correct predictions made
by the classification model as (C11 + C00) and the total number of incorrect
predictions as (C10 + C01).
A good classification model is expected to have more records in cells C11
and C00 and fewer records in C01 and C10. The most popular performance
metric for evaluating the merit of a classifier is the accuracy, defined by:
Number of correct predictions
Accuracy =
Total number of predictions

C11 + C00
=
C11 + C00 + C10 + C01

Similarly, to find the error rate of the classification model, we use the follow-
ing equation:
C01 + C10
Error Rate =
C11 + C00 + C10 + C01
8 Machine Learning for Healthcare

TABLE 1.3
Supervised Algorithms
Algorithm Type Description
Linear regression Regression This technique correlates each feature to
the output which helps to predict
future values.
Logistic regression Classification This technique is an extension of linear
regression, used for classification tasks,
and takes its output variable as binary.
Decision tree Regression/Classification It is a model which is used for
predicting the values by splitting the
nodes into the children nodes by
forming structure of a tree.
Support Vector Regression This algorithm is best used with a
Machine (SVM) non-linear solver. It catches a hyper
plane to distribute the classes optimally.
Naïve Bayes Classification or Regression Naïve Bayes classification technique
finds the prior knowledge of an event
in relation to the independent
probability of each feature.
AdaBoost Classification or Regression It uses a mass of models to have a
decision which weighs them based on
the accuracy in prediction.
Random forest Classification or Regression Random forest uses the “majority vote”
method on multiple decision trees to
label the output.
Gradient-boosting Classification or Regression It focuses on the error generated by the
preceding trees to update the results.

TABLE 1.4
Confusion Matrix for a Binary
Classification Problem
Predicted Class
Actual Class 1 0
1 C11 C10
0 C01 C00

TABLE 1.5
Confusion Matrix for a Three-Class
Problem
Predicted Class
Actual Class C1 C2 C3
C1 C11 C12 C13
C2 C21 C22 C23
C3 C31 C32 C33
Fundamentals of Machine Learning 9

The key objective of a classification model is to find the highest accuracy and
lowest error rate.

1.5.1.1 Supervised Learning Use Cases

1. Cortana: this automated speech system is used for mobile applica-

tions. First it trains itself by using mobile phone voices and then it
makes predictions based on this data.
2. Weather Apps: weather apps are used to predict future weather by
exploring the conditions for a given time, based on previous data.
3. Biometric Attendance: the machines can be trained with inputs of
biometric individuality which can be iris, thumb, or earlobe, etc.
After training, the machine can easily identify the person.

1.5.2 Unsupervised Learning
No labeled data is provided in this type of learning. The algorithm finds the
patterns within the dataset and acquires them. The algorithm clusters the data
into various groups based on its concreteness. By this means, one can accom-
plish visualization of high dimensional data. In unsupervised learning we do
not give goals to our model through training. The model needs to discover what
direction it can realize without anyone else’s input. Figure 1.5 shows the data-
set which contains the data of clients that subscribe to an organization “abc.”
When buying in, they are given a participation card; thus the shopping center
has total data about each client and all his/her purchases. Presently, by utilizing
this information and solo learning strategies, shopping centers can undoubtedly
amass customers, dependent on the parameters they are bolstering in.
Training data is:

• Unstructured data
• Unlabeled data

FIGURE 1.5
Data for unsupervised learning.
10 Machine Learning for Healthcare

Data scientists prefer to choose unsupervised learning techniques over

supervised learning. Some of the primary reasons for this are: i) this tech-
nique finds all kinds of unknown patterns in data; ii) it is easier to get
unlabeled data than labeled data (hence the process of data collection is
easier); iii) these methods help to find features which can be useful for
categorization.

1.5.2.1 Types of Unsupervised Learning

• Clustering: comprehensively, this method is useful for grouping

information dependent on various examples. For instance, if we are
not given a response parameter as shown in Figure 1.6, then this
system will be utilized to amass customers dependent on the infor-
mation parameters given by available information.
• Association: this is a rule-based on a machine learning procedure
which discovers some extremely valuable relationships between
parameters of a huge informational index. Association rules per-
mit us to generate associations in the midst of data objects inside
enormous databases. This unsupervised technique is related to
determining attention-grabbing relationships between variables
in large databases. Generally, most of the machine learning algo-
rithms work with numeric datasets and therefore tend to be math-
ematical. However, association rule mining is appropriate for
non-numeric, categorical data, and needs impartial – more than
simple – counting.

FIGURE 1.6
Clusters of data.
Fundamentals of Machine Learning 11

1.5.2.2 Clustering
Clustering is an unsupervised learning technique in which class labels are
not provided in test samples. It is the process of grouping a set of objects
into classes of similar objects. Collection of similar data objects is known as
a cluster, as shown in Figure 1.6. Similar objects are grouped into one clus-
ter. Clustering is also named data segmentation because it partitions large
datasets into groups as per the similarity of records. Clustering-based pro-
cesses are adaptable to changes and help to draw features that are used to
distinguish the different groups. To determine the similarity between two
data objects, a metric from the datasets (distance function) is used by the
clustering technique. This distance function takes two objects as its input,
and returns the distance between these two objects as its output in the form
of a real number. A smaller value of this real number represents that two
objects are more similar, as compared to a larger value. The goal of cluster
analysis is to group data objects based on the information related to data
which describes the objects and their relationships. In practice, various types
of clustering techniques are used. A partial clustering technique divides sets
of data objects into clusters in a manner that means each object belongs to
a single cluster only, whereas in hierarchical clustering, nested clusters are
organized as a tree. The root of the tree is the master cluster, which contains
all the data objects, and each node in the tree represents the union of its
children. In some situations, a point may be placed in more than one clus-
ter. Such overlapping or non-exclusive clustering represents that an object
simultaneously belongs to more than one cluster. For example, the person
may be a patient as well as employee of the hospital at the same time. In
fuzzy clustering, some membership weight is associated with every object
that represents the probability of belonging to a cluster. Such clusters are
treated as fuzzy sets. Another distinction between clustering techniques is
complete and partial clustering, where each object is assigned to a cluster in
complete clustering, whereas this does not happen in partial clustering. In
clustering, the usefulness of a cluster is defined by the goals of its analysis.
Therefore different kinds of clusters used are well separated into prototype-
based, graph-based, density-based, and conceptual clusters. Based on this,
various commonly used clustering techniques are the Partitioning Method,
the Hierarchical Method, the Density-Based Method, and the Grid-Based
Method.
The major applications of clustering include targeting similar people and
deciding on things like the location for an activity (i.e., exam centers), the
location for a business chain, or for planning a political strategy.

1.5.2.3 Association Rule
Association rule mining is also known as “market basket analysis”.
Association rule mining is a technique which aims to perceive frequently
occurring patterns, correlations, or associations in datasets of various types
12 Machine Learning for Healthcare

of databases such as relational databases, transactional databases, and other

forms of repositories. An association rule comprises of two parts – “if”, in the
form of the antecedent, and “and then”, in the form of the consequent.
“If a customer buys shoes, he’s 70% likely to buy socks.”
In the above association rule, shoes are the antecedent and socks are the
consequent. Figure 1.7 represents the process of association rule mining. In
real life, the number of frequent item-sets is large, which results in a large
number of association rules for a transactional database. The discovery of
frequent item-sets with item constraints is therefore an important problem
in association rule analysis.
Association rules are generated by meticulously evaluating data and look-
ing for frequent if/then patterns. Based on the succeeding two parameters,
these significant relationships are witnessed:

1. Support: indicates how often the if/then association appears in the

database.
2. Confidence: expresses the number of times these relationships have
been found to be true.

Some of the examples of association rules are:

1. Market Basket Analysis: this is the most classic example of associa-

tion mining.
2. Medical Diagnosis: association rules in medical diagnosis can be
useful for assisting physicians to cure patients. Diagnosis is not an

FIGURE 1.7
Process of association rule mining.
Fundamentals of Machine Learning 13

easy process and has a scope of errors which may result in unreli-
able end-results.
3. Census Data: this application of association rule mining has vast
potential in associating sound public policy and carrying forth com-
petent functioning of a democratic society.
4. Protein Sequence: proteins are sequences made up of 20 types
of amino acids. The dependency of the protein functioning on its
amino acid sequence has been a topic of prodigious research.

1.5.2.4 Unsupervised Learning Use Case

A companion welcomes “you” to his gathering where “you” meet total
strangers. At that point, you will order them utilizing unsupervised learning
(no earlier information) and this characterization can be based on age, dress,
sexual orientation, instructive capability, or any other attribute you wish.
Because you didn’t have any earlier information about these individuals, you
simply grouped them “on the go”.
Table 1.6 illustrates the difference between supervised and unsupervised
learning.

1.5.3 Reinforcement Learning (RL)

Reinforcement Learning (RL) is about identifying the proper action to exploit
return in a certain situation, employed by numerous machines to find the
best possible behavior they should take in a distinct state. It varies from
supervised learning in such a way that in supervised learning the label key
lies with the training data, so the model is trained with the correct answer
itself, whereas in RL no label is provided but the reinforcement agent adapts
to accomplish the particular task.
An example can be stated as follows:
There are an agent and a reward, with many hurdles stuck between. The
role of the agent is to find the most promising path to get the reward. This
can be seen more clearly in Figure 1.8 below.

TABLE 1.6
Supervised versus Unsupervised Machine Learning
Parameters Supervised machine learning Unsupervised machine learning
Input Data Training of algorithms is through Unlabeled data is used directly by the
labeled data. algorithms.
Computational Supervised learning is easy to Computationally complex.
Complexity understand and implement.
Accuracy This is a more accurate and Comparatively not as accurate and
reliable method. trustworthy a method, but better for
bigger data sets.
14 Machine Learning for Healthcare

FIGURE 1.8
Reinforcement learning.

Figure 1.8 shows a diamond, a robot, and fire. The robot’s job is to get the
reward in the form of the diamond by avoiding the hurdle in the form of fire.
The robot explores all possible paths, and then chooses the path with least
hurdles based on its learning. Each right step will offer the robot a reward
and each wrong step will subtract the reward. At last the total reward is
calculated.

1.6 Conclusion
Machine learning is an innovative technology that is currently in a serious
phase of several growing and established industries. Machine learning per-
sonalization algorithms are used to distribute recommendations to users
and tempt them into an array of positive actions. Presently, many of such
recommendations are erroneous and frustrating, which hampers users’
familiarity – though in the future the personalization of algorithms is likely
to be tweaked, leading to far more beneficial and successful experiences.
Developers will be able to shape more fetching and discoverable applica-
tions that can commendably understand users’ needs based on natural com-
munication techniques. We presume that, shortly, robots will become extra
intelligent and capable of finishing tasks. Drones, robots in manufacturing
places, and other types of robots are likely to be increasingly used to make
our lives easier (Figure 1.9).
Machine learning is one of the most disruptive technologies of the 21st
century. Although this technology can still be considered as nascent, its
Fundamentals of Machine Learning 15

FIGURE 1.9
Reinforcement learning example.

future is bright. In the coming years, we are likely to see more advanced
applications – one that stretch machine learning’s capabilities to unimagi-
nable levels.

References
Agrawal, Rashmi. 1–50. http://hdl.handle.net/10603/169657. https://shodhganga.i
nflibnet.ac.in/handle/10603/169657.
​ ​ ​ ​ ​ ​ ​
Agrawal, Rashmi. “Design and development of data classification methodology for
uncertain data.” Indian Journal of Science and Technology 9(3) (2016): 1–12.
Agrawal, Rashmi. “Integrated effect of nearest neighbors and distance measures in
k-NN algorithm.” In: Big Data Analytics, Aggarwal V., Bhatnagar V., Mishra D.
(Eds). Springer, Singapore, 2018: 759–766.
Agrawal, Rashmi. “Integrated parallel K-nearest neighbor algorithm.” In: Smart
Intelligent Computing and Applications, Satapathy S., Bhateja V., Das S. (Eds).
Springer, Singapore, 2019: 479–486.
Agrawal, Rashmi. “Technologies for handling big data.” In: Handbook of Research on
Big Data Clustering and Machine Learning, Fausto Pedro Garcia Marquez (Ed). IGI
Global, 2020: 34–49.
Agrawal, Rashmi, and Babu Ram. “A modified K-nearest neighbor algorithm to
handle uncertain data.” 2015 5th International Conference on IT Convergence and
Security (ICITCS). IEEE, Malaysia, 2015.
Batra, Mridula, and Rashmi Agrawal. “Comparative analysis of decision tree algo-
rithms.” In: Nature Inspired Computing, Panigrahi B., Hoda M., Sharma V.,
Goel S. (Eds). Springer, Singapore, 2018: 31–36.
16 Machine Learning for Healthcare

El Naqa, Issam, and Martin J. Murphy. “What is machine learning?” In: Machine
Learning in Radiation Oncology, El Naqa I., Li R., Murphy M. (Eds). Springer,
Cham, 2015: 3–11.
Han, Jiawei, Jian Pei, and Micheline Kamber. Data Mining: Concepts and Techniques.
Elsevier, 2011.
Hastie, Trevor, Robert Tibshirani, and Jerome Friedman. The Elements of Statistical
Learning: Data Mining, Inference, and Prediction. Springer Science & Business
Media, 2009.
Hawkins, Douglas M., Subhash C. Basak, and Denise Mills. “Assessing model fit
by cross-validation.” Journal of Chemical Information and Computer Sciences 43(2)
(2003): 579–586.
Kim, Ji-Hyun. “Estimating classification error rate: Repeated cross-validation,
repeated hold-out and bootstrap.” Computational Statistics and Data Analysis
53(11) (2009): 3735–3745.
Neelamegam, S., and E. Ramaraj. “Classification algorithm in data mining: An over-
view.” International Journal of P2P Network Trends and Technology (IJPTT) 4(8)
(2013): 369–374.
Raschka, Sebastian. “Model evaluation, model selection, and algorithm selection in
machine learning.” arXiv Preprint ArXiv:1811.12808 (2018).
Samuel, Arthur L. “Some studies in machine learning using the game of checkers.”
IBM Journal of Research and Development 3(3) (1959): 210–229.
Simeone, Osvaldo. “A brief introduction to machine learning for engineers.”
Foundations and Trends® in Signal Processing® in Signal Processing 12(3–4) (2018):
200–431.
Stamp, Mark. Introduction to Machine Learning with Applications in Information Security.
Chapman and Hall/CRC, 2017.
Tarca, A. L., V. J. Carey, X. W. Chen, R. Romero, and S. Drăghici. “ML and its applica-
tion in biology.” PLOS Computational Biology 3(6) (2007): 116.
Witten, Ian H., and Eibe Frank. “Data mining: Practical machine learning tools and
techniques with Java implementations.” Acm Sigmod Record 31(1) (2002): 76–77.
Zheng, Alice. “Evaluating Machine Learning Models: A Beginner’s Guide to Key
Concepts and Pitfalls.” (2015).
2
Medical Information Systems

Uday Sah, Abhushan Chataut, and Jyotir Moy Chatterjee

CONTENTS
2.1 Introduction................................................................................................... 17
2.2 Types of Medical Information Systems..................................................... 18
2.2.1 General Medical Information Systems.......................................... 18
2.2.2 Specific Medical Information Systems........................................... 19
2.3 Types of General Medical Data................................................................... 19
2.3.1 Numerical Data................................................................................. 19
2.3.2 Textual Data....................................................................................... 20
2.3.3 Categorical Data................................................................................ 20
2.3.4 Imaging Data..................................................................................... 20
2.4 History of Medical Information Systems.................................................. 20
2.5 Collection of MIS Data through Various Platforms................................. 21
2.5.1 Traditional..........................................................................................22
2.5.2 Electronic...........................................................................................22
2.6 Diagnosis and Treatment of Disease through MIS Data........................22
2.7 Conclusion..................................................................................................... 23
References................................................................................................................ 23

2.1 Introduction
Medical information systems (MIS) are the universal purpose of computer-
based data storage of disease diagnosis and related medical data that help
investigate medical cures and other patient information (Johnson and
Barnett 1977). The major purpose of MIS is to help the decision-maker (doc-
tor or machine) find the most appropriate medicine and track patient health-
related information. It could happen in real-time or could be from historical
data. In the current era, millions of people are dying due to lack of a medical
information system that can hold patient data and give insights form the
data to make automation in healthcare industries possible. Medical infor-
mation system stores various kinds of medical data like blood group, sugar
level, and much more, to identify and easily recognized disease through
symptoms.

17
18 Machine Learning for Healthcare

MIS data are widely accepted for analytical use in artificial intelligence
research to discover unique patterns of disease symptoms which may lead
of the invention of a cure.
MIS data are stored in file orientation data structures that can easily be
searched by doctors to find the major disease symptoms of a patient. This
technology will rapidly grow in the future because the needs of scientific
sampling and analysis for improved drug discoveries requires it. Medical
information like this is important in terms of the clinical inspection and
evaluation of disease. MIS can generate huge amounts of statistical data that
require capable data management infrastructures for storage, retrieval, and
management according to doctors’ needs (Johnson and Barnett 1977).
For MIS information security is highly concern, because it moves patients’
critical clinical information between different computers (Brook 2019). This
information is sent through electronical channels. This method helps the
many departments of hospital operational management to reduce their paper
usage in prescriptions given by doctors as part of test information, as well as
helping eliminate incorrect clinical information. MIS is integrative in nature
for clinical data storage so it could be helpful in emergency situations when
patients are unconscious and unable to perform the tests required by doc-
tors. When this happens, the ability to retrieve their historical data can be
important to safely treat or give an initial diagnosis of the patient. Most of
the MIS and medical investigative research is based on unbiased data that
are collected through various sensors and medical devices. In the current era,
hospitals are more responsible when conducting medical research and col-
lecting data in purist of making useful artificial intelligence applications that
can save billions of people from deadly diseases. That kind of application
interprets data in a such a way so as to predict deadly diseases in early stages.

2.2 Types of Medical Information Systems

There are several types of medical information systems that make it possible
to manage huge amounts of health data for both general and specific pur-
poses relating to the prevention and prediction of noxious diseases. Medical
information systems can perhaps be defined according to two categories, but
there are many other perspectives.

2.2.1 General Medical Information Systems

General medical information systems store basic information which relate to
the disease or diagnoses of patients. Traditional electronic medical record-
ing systems are a valuable tool for providing knowledge that can serve as a
reference for quality decision-making in prescriptions (Ting et al. 2003). The
Medical Information Systems 19

systems contain basic monitoring, and up-to-date data collection from patients
to track diagnoses and investigate disease. Examples include blood test infor-
mation, sugar level information and related blood pressure, as well as urine
information – all of which are very important in finding symptoms of under-
lying deadly diseases. They contain more data than found in specific medical
information systems which helps to use the information resources to create
and manage AI data resources. This kind of data demands high levels of clean-
ing and data accuracy for the future AI application of the medical information
system. For example, patient data management and storage systems, etc.

2.2.2 Specific Medical Information Systems

Specific medical information systems are a special category of information
system which store disease-specific data storage. This helps to accurately and
properly diagnose and investigate stages of a disease. The systems contain
various disease-specific data, for example, if a patient has brain cancer or a
tumor the system used will contain only the data which is related to that dis-
ease. It also stores the symptoms and testing information needed for diagno-
sis by the doctors. It is a more disease-oriented medical information system,
one that is precise and effective in finding symptoms and future treatment
which relates to a particular disease. Examples include brain cancer medical
information system, chest pain medical record system, and more.

2.3 Types of General Medical Data

There are various kinds of medical data generated by hospitals on a daily
basis. That kinds of data required for diagnosis complex disease in scientific
research and treatment of patient healthcare industries. The group of medi-
cal scientists have categorized and generalized the medical data in a four
category. These are descripted below:

2.3.1 Numerical Data
Numerical data are the more common data in healthcare industries when mak-
ing measurements of patient-generated data. For example, measurement of
blood pressure, or sugar levels. It may be discrete or continuous in nature. If
brain cancer has four stages, 0 is defined as the absence of brain cancer, 1 is
defined as a low chance of brain cancer, 2 is defined as a medium chance of
brain cancer, and 3 is defined as a high chance of brain cancer, and so on. That
type of data is known as discreate numerical data. Another type is continuous
data which cannot be counted and can only be described through range inter-
vals (Rumsey 2016). For example, blood pressure in the range of 70–120 mmHg
systolic and in diastolic 60–80 mmHg.
20 Machine Learning for Healthcare

2.3.2 Textual Data
Textual data is one of the most popular and well-defined syntaxial word
data for human understanding. It contains alphabetic data. For example, if
the patient likely has heart disease then textual data will be more represen-
tative and effective in order to understand and provide a clear meaning for
the reader to evaluate the words and sentences. Most of the medical data are
textual data which provide clear meanings and disease descriptions relat-
ing to patients. This kind of information is useful when doctors diagnose
and treat diseases. Textual medical data in machine learning can be used
for such things as predicting heart disease in patients through doctors’ pre-
scribed medical report results.

2.3.3 Categorical Data
Categorical data refers to the type of textual data that can specify category
and it is an effective way to divide and categorize when medical data are
scattered. Categorical data are mostly used in categorial machine learning
prediction, which is specifically the level of seriousness of disease – in other
words, dividing patient seriousness into more comprehensible and visual-
ized forms. Categorial data are represented by mathematical expressions
which are not clear meaning but are useable during training and pre-process
of AI application development. For example, blood group categized patient
treatment or drug classification.

2.3.4 Imaging Data
In MIS, imaging data plays as important role in diagnosis and treatment of
disease in hospitals. Imaging data are pixel-based multi-dimensional data
which are most effective while developing image recognition and classifica-
tion models to automatically recognize disease and investigate the serious-
ness of the disease. It also helps doctors to visualize anatomical and internal
parts of the body’s organs and, thus, identify infected areas. Some examples of
imaging data include X-Ray data, ultrasound image data, Medical Resonance
Imaging, and Computerized Tomography scan data. Each piece of data is
stored in various MIS in unistructural databases like MongoDB. For some
reason video data are stored in the form of frame imaging. In MRI data collec-
tion, information about visualized the location of water presence (Klein 2018).

2.4 History of Medical Information Systems

Doctors in the 1500s borrowed from bookkeeping practices and modeled
themselves on Hippocrates when they recorded case histories on clay tablets.
Medical Information Systems 21

Some kept account books written at the time of encounter, while others main-
tained journals or diaries which enabled them to study after a day of visiting
patients. These all typically included date, patient’s name, age, complaint,
possible cause, a prescription, and proof of payment for recording the his-
tory of disease and cure. These were known as “casebooks” which were also
the first age of information overload. This practice evolved from expensive
parchment to more affordable paper technologies to modern-day computers.
From the early 1600s, the presentation of information in medical records led
to the improvement of diagnosing and treating illness, which was through
observation as well as actions in the treatment of patients. This became such
a reputed method that it began to be used to teach other medical practitio-
ners. Paper-based record keeping continued until the 1920s (NCBI 2015). In
the 1960s, the development of computers led to disruption in healthcare and
provided the opportunity to maintain records electronically. However, the
cost of maintaining the mainframe computers and the storage of data led
to only large organizations having full funded use of electronic technology.
In the 1970s, as the size of computers became smaller, software began to be
designed for day-to-day healthcare functions like patient registration, phar-
macy use, and clinical laboratory ad billing but were still based by depart-
ment. The multiple separate system integration was developed in the 1980s as
“hospital information systems” (HIS) which were able to connect the finan-
cial and clinical systems, but the functions were very limited. Advancement
in healthcare technology integration grew in the 1990s and gave access to
computing systems that would share data and information across all the
separate systems. In the 2000s, the importance of integrated electronic health
records for better decision making grew. The electronic health record revolu-
tion began in 2004 leading to an increase in volume of data, accessibility to
data, and management of health data relying on health informatics. From
2010 to present, the context of Health Information Technology focuses on
value-based care over fee-based care and, thus, leads and improves patient
outcomes by harnessing the growing volume, variety, veracity, and veloc-
ity of healthcare data to support clinical (as well as operational) decisions
in healthcare. Many learning tools, such as machine learning, help improve
patient outcomes by aiding healthcare professionals to improve care on both
broader and larger scale.

2.5 Collection of MIS Data through Various Platforms

MIS data are collected by various platforms and sources that help data
engineers to store the data in structural and unistructural databases. For
example, MIS data are collected through heath record systems which are
manually entered by hospital staff and employee. It can also be collected
22 Machine Learning for Healthcare

through human wearable devices. For now, we are going to divide medical
data collection into two approaches. The first is Traditional and the second is
Electronic. There are discussed in detail below.

2.5.1 Traditional
In the traditional method, patients’ diagnoses details are stored in an MIS
database after the hospitals; staff manually enter and store the data in an
MIS system. Most hospitals are tradition-based thus they store test results
by manually entering them into an MIS system. For example, hospital staff
manually store patient details like name, gender, age, blood group, etc. The
traditional method is one of the oldest methods of storing medical data, but
it is a less accurate method for medical data collection and less useable for
medical scientific research. In the traditional method most hospital work is
still paper-based. Staff are busy and under too much time pressure which
leads to patient data collection that is less accurate and beneficial for medical
scientific research.

2.5.2 Electronic
Electronically collected data is a modern method of medical data collection.
It is more accurate and up-to-date, real-time collected patient data that uses
techniques which see data collected through various forms of electronic sen-
sors. Nowadays, most countries have commercialized IoT devices for public
use and this helps us save time in the collection of patient data by utiliz-
ing wearables. The latest wearable devices are well-equipped for collecting
patient medical data and notifying both doctors and hospitals when a patient
is in a critical situation. Electronic sensors are capable of sensing a patient’s
health status and directly sending doctors a portal to provide emergency
help to the patient. Many current MIS are purely based on modern electronic
sensor-based data collection for hospitals to recognize, diagnosis, and cure
patients’ diseases. Electronic-based medical data recording is widely accept-
able for machine learning, and deep-learning activities for making better
and more useful AI applications. Nowadays, big data are evolving so that
data analytics can develop the innovative technology to support billions of
patients receiving healthcare. This technology is capable of solving inaccu-
rate medical data issues.

2.6 Diagnosis and Treatment of Disease through MIS Data

Diagnosis and treatment are essential roles of MIS data for curing deadly
diseases in medical science. Medical big data are used in AI applications to
Medical Information Systems 23

make systems that can evaluate and that are self-able to provide medication
appropriate to a patient’s illness. In the current era, most medical data are
alphanumeric or imaging data. THIS is helpful for doctors so they directly
can diagnosis patient disease with the help of patient check-up reports.
Diagnosis has important implications for medical science and patient care,
for disease research, and for treatment which can all use checkup medical
data. (Balogh et al. 2015).
To understand the diagnosis process, an example is outlined here: Once
a patient suffers a health problem, the patient or his/her family gather any
possible information about the health problem and collect the symptoms to
easily extract data and explain what they are suffering from. Next, after con-
tact with clinical hospitals for diagnosis and treatment about facing health
problems, doctors see the types of symptoms and details to identify pos-
sible diseases and extract the data according to the information given by
patients. Doctors are responsible for explaining to patients about why they
facing health problems, and advising them on the best ways to overcome
those health problems. Nowadays, a large amount of health activities and
their symptoms can be collected through wearable devices which are easily
connected to an MIS system. When doctors want to check an MIS, they can
easily receive the data as well as information that can be used to provide
clinical medication to the patient. When doctors communicate with patient,
they gather information.

2.7 Conclusion
MIS offers many benefits over the executive systems that came before. They are
a completely operational, online system, which permit incredible adaptabil-
ity for information depiction and examination. They can be executed on a PC
system of moderate expense. Permissible information types include numeric,
categoric, content, and date. Each field may have one or many sections. It is
easy to imagine a few arrangements of comparable information, including
date-situated information, to be gathered for every person in the investigation.

References
Balogh, E. P., Bryan T. Miller, and John R. Ball. 2015. Improving Diagnosis in Health
Care. Washington, DC: National Academies Press (US).
Brook, C., 2019. What Is a Health Information System? [Online]. Available at: https​://
di​g ital​g uard​i an.c​om/bl​og/wh​at-he​a lth-​i nfor ​m atio​n-sys​t em. [Accessed 6
January 2020].
24 Machine Learning for Healthcare

Committee on Diagnostic Error in Health Care; Board on Health Care Services;

Institute of Medicine; the National Academies of Sciences, Engineering, and
Medicine, E. P. Balogh, B. T. Miller, J. R. Ball, editors. 2015. Improving Diagnosis
in Health Care. Washington, DC: National Academies Press (US), December
29. 2, The Diagnostic Process. Available from: https​://ww​w.ncb​i.nlm​.nih.​gov/
b​ooks/​NBK33​8593/​.
Johnson, D. C., G. O. Barnett. 1977. Medinfo - A Medical Information System.
Medinfo - A Medical Information System, pp. 191–201.
Klein, C., 2018. Intro to Analyzing Brain Imaging Data— Part I: fMRI Data Structure
[Online]. Available at: https​://me​dium.​com/c​oinmo​nks/v​isual​izing​-brai​n-ima​
ging-​data-​fmri-​with-​pytho​n-e1d​0358d​9dba [Accessed 27 Jan 2020].
Rumsey, D. J., 2016. Types of Statistical Data: Numerical, Categorical, and Ordinal
[Online]. Available at: https​://ww​w.dum​mies.​com/e​ducat​ion/m​ath/s​tatis​tics/​
types​-of-s​tatis​tical​-data​-nume​rical​-cate​goric​al-an​d-ord​inal/​. [Accessed 23 Jan
2020].
Ting, S. L., W. H. Ip, A. H, Tsang, G. T Ho. 2003. An Integrated Electronic Medical
Record System (iEMRS) with Decision Support Capability in Medical
Prescription. An Integrated Electronic Medical Record System (iEMRS) with
Decision Support Capability In Medical Prescription, pp. 236–245.
3
The Role of Metaheuristic
Algorithms in Healthcare

G. Uma Maheswari, R. Sujatha, V. Mareeswari, and E. P. Ephzibah

CONTENTS
3.1 Introduction................................................................................................... 25
3.2 Machine Learning in Healthcare............................................................... 26
3.3 Health Information System Framework.................................................... 27
3.4 Privacy and Security of Data...................................................................... 28
3.5 Big Data Analytics in Disease Diagnosis.................................................. 29
3.6 The Metaheuristic Algorithm for Healthcare........................................... 32
3.7 Conclusion..................................................................................................... 35
References................................................................................................................ 36

3.1 Introduction
The process of machine learning is known as the “training” of machines
and the output generated is known as the “model”. Machine learning is a
data analytics tool which automates the building of analytical models. It is a
system that gains knowledge from data, with minimal human intervention.
Data is provided to the algorithm and it in turn creates new knowledge from
what it has learned before. The classification model is meant to assess a cat-
egory as “it’s one thing or another”. The model is educated during the catego-
rization of the dataset. The clustering model is generated when a bunch of
data is available but has not defined a result and simply wants to see distinc-
tive trends in the data. The regression model is developed for value detection
purposes. The algorithm will find a correlation between two variables, using
data, and the result is predicted accordingly. Healthcare providers are able
to provide better decisions about treatment options for patients, leading to
overall changes in healthcare services with the help of machine learning.
Machine learning algorithms are also useful in providing medical doctors
with broad statistics of real-time data and advanced analyses about the con-
dition of the patient, laboratory test outcomes, blood pressure, family history,
clinical trial data, etc. The new aims of healthcare is preventative – to avoid

25
26 Machine Learning for Healthcare

illness at the early stages instead needing to go for assessment and diagnosis
for treatment. Nevertheless, new technological developments like big data
and machine learning provide more precise results that can be obtained for
predicting disease. When machine learning models are exposed to raw data
the iterative nature enables the model to change independently.

3.2 Machine Learning in Healthcare

Today, machine learning is helping to standardize hospital operating pro-
cedures. This also facilitates bacterial infection control and diagnosis, and
also customizes medications. The rate of change in technology is improv-
ing drastically, which is reflected in healthcare sector as well. The fields of
statistics, probability, and Artificial Intelligence comprise the fundamental
machine learning subjects. In earlier medical treatments there were many
manual documents and procedures which were very tedious. In technology,
there are improvements for maintaining electronic medical records which
are more efficient when compared to the manual records maintained ear-
lier. In order to improve healthcare in the future, computational capacity and
machine learning need to strengthen the electronic information given to doc-
tors. By using this sophisticated kind of research, at the point of patient care,
we can provide doctors with better information. Physicians need to be pro-
vided with more knowledge so they can act independently and make better
decisions regarding patient diagnoses and choice of treatment, while recog-
nizing the potential outcomes and costs for each. The importance of machine
learning in healthcare lies in its ability to process large datasets beyond the
scope of human capability, and then efficiently turn data analysis into clini-
cal insights that help doctors to prepare and provide treatment, eventually
leading to better outcomes, lower cost of care, and increased patient satisfac-
tion. Machine learning can be used to train computers to see images, identify
abnormalities, and point to areas that need attention – thus enhancing the
accuracy of all of these processes. Long-term, machine learning may also
support family practitioners. To improve further efficiency, reliability, and
accuracy, machine learning may give objective opinions.
As has been mentioned, in healthcare the best method of machine learn-
ing is the doctor’s brain. Doctors may be afraid that machine learning is the
beginning of a phase that could make them redundant. This fear will be
reduced by both machine learning and other future technologies in medi-
cine. Machine learning is considered as a tool to be used by physicians to
boost treatment. Different use cases must be identified in which machine
learning technologies provide value from certain technology applications.
This will be a step towards more incremental processing for more analytics,
machine learning, and prediction algorithms to be incorporated into daily
Metaheuristic Algorithms  27

clinical practices. Medicine does have a system for researching and demon-
strating safe and effective treatment. This is a prolonged trial process and
is based on evidence to support the decision. When we consider machine
learning, it needs to be ensured that that same process is in place to guar-
antee safety and effectiveness. We must consider the ethics of giving away
some of the things we do to a machine. Many people may wonder if this is
just a “fad” in technology, or if it does bring real value to healthcare. Health
Catalyst believes that it is one of the most significant life-saving innovations
ever developed and there will soon be the implementation and widespread
use of machine learning in healthcare. We assume that prospects for clini-
cal change and growth are not in fact limited to technology; reacceptance in
machine learning is reduced in a guided, accurate, and patient-centric way.
Doctors will receive daily advice on how to reaccept patients and reduce the
risks. Through machine learning, a hospital system can recognize chronic
diseases of patients who are undiagnosed. It also provides patients-centered
preventive interventions to avoid the risk of developing chronic disease in
patients.

3.3 Health Information System Framework

Deep learning is a promising end-to-end form of learning for accurate diag-
nosis of health status. There are two main components in an end-to-end
system that emulate the healthcare cycle. The first is the deep recognition
module focused on deep neural networks used to make health diagnoses.
The second module is based on Bayesian inference graphs to determine the
operation (Dai and Wang 2018). Cloud computing has taken on an important
role in healthcare services too because of its ability to improve the efficiency
of healthcare services. Optimum virtual machine selection achieves a signif-
icant performance enhancement by reducing the execution time of medical
requests from stakeholders, and optimizing cloud resources utilization. For
this, the virtual machine selection is optimized by a new model based on the
cloud environment using a parallel Particle Swarm Optimization (Abdelaziz
et al. 2018). Automatic machine learning – or auto AML – is a field that’s been
gaining a lot of popularity. Whenever we have a dataset and want to collect
insights from it using machine learning there are a whole lot of steps that
we have to complete, such as cleaning the data, selecting the most relevant
features, and choosing the right model. This process can be both long and
expensive because it requires a lot of guessing and checking to ensure we
have the best possible results. The goal of auto AML is to automate as many
of these steps as possible without compromising the accuracy of our results
(Christopher 2019). Large volume of complex data has always entered the
world of healthcare, flowing in at a very rapid pace. Across various sectors
28 Machine Learning for Healthcare

of the healthcare industry, a huge volume of data is generated by hospitals,

the healthcare sector, medical devices, medical insurance, and research.
Technological advancements enable the use of large volumes of data to
improve the healthcare sector. By applying analytics, machine learning and
big data allow trends and associations to be found and thus offer actionable
insights to improve healthcare delivery (Dai and Wang 2018). In healthcare,
a “database” refers to the health information system. Health management
systems include the options to add, delete, modify, and access records of
patients, as well as the operational management of a hospital system that
supports policy decisions on healthcare (Hegde and Rokseth 2020).

3.4 Privacy and Security of Data

Every time a new patient is admitted to hospital, a new medical record
is created. The medical record documents relevant information about the
patient during their stay. It forms the legal record of care provided and is
essential for clinical communication when the patient is discharged. The
medical record is also sent to the hospital coders who take the information
and convert it into code. These codes are then used to identify trends and
patterns of disease, improve the quality of safety and care, and analyze the
cost of providing quality care so that appropriate funding for healthcare
services can be received. An incomplete or ambiguous medical record can
lead to inaccurate coding. Due to strict coding rules, codas cannot make
assumptions about diagnoses, procedures, medications, and complica-
tions. If a medical record is unclear, hospitals and their patients can miss
out on the funding they are entitled to. If the details aren’t clearly written
down, it may as well not have happened. To avoid this, clinicians should
ensure that clinical notes are complete, legible, and updated in a timely
manner. They should clearly document the principal diagnosis as well as
any additional diagnoses. The necessity of clinical records is to provide
and compute clear information about symptoms and the medical treatment
provided to patients and identify all diagnostic and therapeutic interven-
tions. Researchers and planners use these clinical records, since clear and
complete information about every patient improves outcomes and the over-
all healthcare system.
When we look at data analytics for medical research, we can break it up
basic categories: genomics, imaging, clinical information, artificial intelli-
gence, and deep learning type workflows. The problem is all of this data is
typically not connected together, they are not able to be communicated. This
requires more infrastructure, key provisions, storage, and then it is neces-
sary to move and copy all of that data from different repositories into this file
system. It is very time consuming and it wastes resources.
Metaheuristic Algorithms  29

A better way is to create a data ocean, where all of that data is taken
and put into one single place so that it’s easier to access. It enables Watson,
Docker, and Power and all these other frameworks that may well be used
on a daily basis, or perhaps not. This framework allows the use of all your
applications and even the new ones that will soon be released that we don’t
even know exist yet because they’re being newly created by researchers
daily. The next step is to stick this orchestration layer in between. This will
allow you to use the entire infrastructure across all of your computer and
all of your applications and workflows, allowing you to most efficiently use
this entire infrastructure. Healthcare organizations are facing the problem
of both patient data security and the need for productivity. As data vulner-
abilities increase, the healthcare organizations are looking for more solutions
in order to resolve these problems.

3.5 Big Data Analytics in Disease Diagnosis

Disease diagnosis is becoming a more challenging task as types of possible
illnesses increase due to the lifestyle changes and modern work environ-
ments that pose great hurdles in the daily life of an individual. However, the
medical field has also evolved a lot in last few decades with the intervention
of Information Technology as a support to speed up the diagnosis process.
Numeric data (generated from the various tests like blood test, urine tests,
thyroid function tests) as well as image and text data (generated from CT
scans, MRI scans, X-Rays, and more) are still used for the purpose of diag-
nosis. The recommendation system is built with the big data received from
various sources and machine learning algorithms are deployed to get the
greater insights. Big data is deployed in the cloud environment and utilized
based on requirement. The advent of big data and usage of cloud comput-
ing greatly reduced the expense of infrastructure investment. Accuracy,
execution time, and robustness metrics are used to gauge the performance
of developed recommender systems that help in diagnosis along with treat-
ment. The ontology constructed for each specialization aids in the clarity of
the ongoing work. Medical ontology is evolving, certainly more than in any
other field. Healthcare has begun to rely on the latest trend of deep learning,
algorithm subset of machine learning, adding strength to the developing
model that supports the various departments of healthcare. For any system
to be reliable, a primary criterion is the intelligent creation of a database that
supports the versatile data. Clinical data is required to be designed with
proper analysis by experts. Huge amounts of work is required to provide
stable data at the backend, and over that the rest of the work begins. Perfect
features are required for the modeling of highly informative, human under-
standable system. Data preprocessing plays a vital role because the collected
30 Machine Learning for Healthcare

data is highly diverse. It may be doctors’ notes, or laboratory reports in the

form of numeric, text or, image. The flow of deploying machine learning
begins with defining the clinical problem and requires proper justification,
at the same. Extracting the data, selection, and refining the same for the scope
point of view decides the structure of system. Analyzing the data, and con-
structing the model, along with its validation ensure it is working correctly.
The human interaction machine will go through a number of trials before it’s
permitted to start its real work in the clinical environment. Neural networks
are a form similar to the neuron structure of a human being and are part of
evolutionary algorithms. This means that multiple layers, as hidden layers,
define the principle of deep learning and enable it to work on huge versatile
amounts of data cumulated from the healthcare system. Data preprocessing
involves labeling, curating, and mapping with justification of each attributes
used in system, such as storing proper indexing and hashing techniques
for faster access based on the intention. These processes adds weightage for
retrieving algorithms effective for study of the output or requirement of the
system. This helps as the deciding factor is front end – proper designing –
and training is vital in creating the best system (Chen et al. 2018; Makwakwa
et al. 2014; Nilashi et al. 2017).
Machine learning is applied in the early detection of Parkinson’s disease
and plenty of work is being carried out to help needy patients at the earli-
est stage. Artificial Intelligence has begun entering the department of diag-
nosis and patient monitoring. Based on the gathered data points from the
patient decision about readmission and attention requirement are tracked.
Information and communication technology scholars across the world have
started working with various disorders like autism, Parkinson’s, the heart
and thyroid, retinopathy, diabetics, and cancer alongside medical experts to
help patients. Based on the data both supervised and unsupervised algo-
rithms are used for the purpose of analysis. Classification and association
rules are used in the categorical dataset. Clustering and regression are used
often in the time series and historical datasets, to form clusters that pro-
vide insight about the nature of the data. Data attributes are the key factors
that make the research or findings so interesting and informative in nature
(Ngiam and Khor 2019; Nilashi et al. 2019; Astrom and Koker 2011).
Devices with various names based on the purpose of diagnosis, detection,
and other healthcare-related terms have begun to emerge. A “point of care”
diagnostic device is the name given to the devices that emerged in the clini-
cal sector as a result of the digital era. Reaching the conventional laboratory
for the preliminary diagnosis requires more time to reach particularly from
very remote area. During an epidemic or acute disease times of high serious-
ness, patients require persistent monitoring that can be achieved with this
point of care device. These devices are a result of an of the Internet of Things
(IoT), along with handling voluminous, versatile data points with compact
algorithms. The Internet of Things and its interconnection of various sen-
sors and devices can remotely coordinate the patient alone with its expertise.
Metaheuristic Algorithms  31

Various sensors are advancing the medical field, including for measuring
temperature, checking blood pressure, and inner organ issues. The research
in this field looks towards incorporating Artificial Intelligence that helps
with diagnosis in non-invasive mode. Point of car diagnostic devices in the
initial stage concentrated only on collection of data, but the mobility and scal-
ability of the collected data required it to be provided at the correct location
to convert it into treatment. Architecture and multiple layers were devised to
substantiate its data flow and analysis (Nilashi et al. 2018; Amirian et al. 2017;
Peeling 2015; Urdea et al. 2006).
Image analysis and video analysis is a research area in the medical arena
that works towards disease diagnoses in the form of pictures in varied for-
mat. Pixel intensity provides the depth of severity. Preprocessing starts from
acquiring the images, extracting, cleaning, and annotation paves the way for
integration along with representation (Wong et al. 2019). Cancer is a com-
mon dreadful disease that threatens people across the globe. Oncology is
supported by big data that helps in the analysis of various cancers like lung,
breast, prostate, salivary gland, and melanoma. The prevalence of breast can-
cer is very high and earlier detection helps to save lives. Images of the breast
help to identify the severity and stage of the cancer. Thermal cameras are
used to capture the images and the intensity of the colors helps in a cancer
diagnosis (or lack thereof) in a perfect manner. Acquired images are run in
the various algorithms to get clarity on if the cells are benign or malignant
(Thai et al. 2012; 38Willems et al. 2019; Rezk et al. 2017). Biomedical images
within information and communication technology safeguards the lives of
millions of people through early diagnosis and medication (De Ridder et al.
2015). Deep learning is used more nowadays to make use of neural network
structures with many hidden layers to achieve a deep analysis of the cumu-
lated data (Kouanou et al. 2018).
The big data analytics in the healthcare sector are being utilized to frame
various architecture based on the input, process, and output. Extensive
analytics are done by gathering works based on data types such as: clini-
cal, patient, sentiment, administration, activity costs, pharmaceutical, R&D,
and data from other databases. This is all in addition to big data techniques
like modeling, machine learning, data mining, visualization, statistics,
simulation, web mining, optimization methods, text mining, forecasting,
and social network analysis. By using big data analytics, various values are
derived and listed as personalized healthcare automated algorithms that can
support professionals in decision making and novel business models. It also
leads to products that provide services, healthcare information sharing, and
coordination facilities, privacy protection practices, the ability to customize
actions by grouping related populations, and the ability to identify problems
at earlier stages. By matching the data type and big data to values, optimized
decision making is feasible (Lee et al. 2017; Galetsi et al. 2020; Baro et al. 2015;
Berger and Doban 2014). Extensive research work is being carried out on
big data in healthcare to ease the burden on future generations and this is
32 Machine Learning for Healthcare

FIGURE 3.1
Terms of “Big data in the healthcare sector”.

evident in the map that can be derived by mapping with the search word
used in the Scopus repository. Figure 3.1 indicates the various related terms
that illustrate the prevalence of big data in the healthcare sector (Liberatore
and Nydick 2008).
Smart cities are a popular new phenomenon across the globe. Smart health
is a component part of smart city projects and many federal governments
are taking large steps to implement effective healthcare to all its citizens in
extensive manner. Making decisions based on the collected big data, with
the available methods, is a great boost for smart healthcare (vosviewer.com)
(Hashem et al. 2016; Palanisamy and Thirunavukarasu 2019).

3.6 The Metaheuristic Algorithm for Healthcare

Metaheuristic algorithm research is a very innovative approach. It’s suitable
for solving large complex project involving large datasets. Optimal results
are retrieved by proper analysis and application of the algorithm. The evolu-
tionary algorithms developed rely on both the nature and characteristics of
the nature. Based on genetic concepts the genetic programming established
in 1989 by Goldberg. Many metaheuristics that are based on ant colony struc-
ture, honey bee structures, firefly, bat, cuckoo, grey wolf, whale, and so on,
are evolving. Despite this biological and environmental background, some
Metaheuristic Algorithms  33

are generated based on mathematical applications like greedy approach,

the traveling salesman problem, differential evolution, and so on. The lat-
est research on metaheuristics is combining multiple concepts and nam-
ing them as hybrid metaheuristic algorithms. The idea and work are highly
unpredictable in nature because the thought of integrating is very inspir-
ing in nature and the ultimate aim is to get the optimal results (Wills 2014).
Working on metaheuristic algorithms is getting to this phase by considering
various nature-based phenomenon (everything from bees to cows). The truth
behind this is evident when searching “metaheuristic algorithm” in Scopus
with open source access, research articles, and English language journals
in engineering and computer science have produced around 745 articles as
the output. It’s really in the beginning phase and definitely can be applied
in various applications without any doubt. The secret is the correct map-
ping of the required data-based on the problem. The application of meta-
heuristic algorithms is widespread and its uses include: solving problems in
production and planning, internet computing, transportation and logistics,
the engineering and agriculture fields, the healthcare industry, and so on
(Dokeroglu et al. 2019; Juan et al. 2015; Pelteret et al. 2018).
The elderly population is increasing across the globe and in this modern
era lifetime longevity is higher due to a higher level of medical support both
in diagnosis and treatment. The need to take care of elderly people and chil-
dren is very challenging and more effort is required. Due to the advance-
ment and interventions of various fields like data mining, machine learning,
big data, the Internet of Things, data analytics, data science, soft computing,
cloud computing, nature-inspired, and population-based metaheuristic algo-
rithms in healthcare makes the way for researchers to make active contribu-
tions based on real-time applications (Shayanfar and Gharehchopogh 2018).
Typical research issues begin with the way data needs to be cumulated and
analyzed because of the varied type of data developed from text, image, video,
and so on. Smart healthcare is a buzzword that circulates in both developed
and developing countries. Optimizing the existing system is the aim of the
developed countries. On the other hand, in developing countries the vision is
to implement a smart healthcare system for their nation in an affordable man-
ner with the support of the researchers. Metaheuristic algorithms possess a
high potential to solve data mining problems like classification, clustering,
and association rule. When compared with the neural network concept and
exhaustive searching concepts, the processing time and solutions obtained in
the metaheuristic approach are much faster and appreciable in nature. The
availability of huge data in healthcare provides a great foundation for com-
puter-based predictions and helps speed up the treatment process. Clinical
prediction has been evolving at an exponential rate over the past few decades
and it’s mainly due to the extensive dataset in the researcher’s hands – the
computational intelligence used in the diagnosis of cardiac disease, cancer,
and so on. The extensive work by utilizing the computation intelligence and
metaheuristic algorithm is carried on for the prostate cancer prediction. This
34 Machine Learning for Healthcare

is gathered based on the collected dataset and is followed by preprocess-

ing, feature selection, feature extraction, and providing the correct sample
to apply the algorithm to get highly accurate results. Cross validation, along
with a proper combination of training and testing, the dataset is the secret
behind the performance of the constructed predictive model (Tsai et al. 2016).
Interesting work has been done on finding illicit pills within the shape
concept. Image-based analyzing is very thought-provoking research in the
medical field and includes x-ray interpretation, scan reports, pill analysis,
and so on. The need for medications has increased and various socio-polit-
ical and economic factors have begun to influence the medical field in the
recent years. Inspection and checking are required and mandated to make
the field safe and ensure the life of the people is not at stake. Through image
processing the quality of pills is segregated (Cosma et al. 2017). Google
trends show ample research work being carried out in metaheuristic algo-
rithms over the past five years across the world. Figure 3.2 illustrates the
trend of metaheuristic algorithms.
The most critical time for a patient is often addressed by the emergency
department of hospitals. A higher level of coordination and commitment
is the main requirement of the people serving in emergency departments.
Part of the study of medicine insists practitioners attend to critical patients
with higher care. Life is the timespan between the birth and death of the
person. Caring for patients and providing satisfactory treatment is the duty
of the healthcare centers. Shortening the waiting time is thus the most press-
ing requirement of any hospital because both caretaker and patient will be
extremely anxious until they meet the practitioner and get support. Work is
done by the genetic algorithm and simulated annealing to optimize the emer-
gency department. It does this by helping make access to the practitioner for
diagnosis purposes faster by scheduling the nurses on duty in the best way
possible (Carneiro et al. 2019). Hospital stays tend to be mandatory in the case
of critical illnesses. Proper treatment along with care both from hospital and

FIGURE 3.2
Trends of metaheuristic algorithms.
Metaheuristic Algorithms  35

patient is key in creating an environment for a the sick person to get better.
After a thorough study of a person’s health, a patient may be discharged by
hospital management after getting input from health experts. The ultimate
aim of any doctor is to ensure the wellbeing of their patient. Longevity of life
is in the interest of all the individuals. Research has addressed the problem of
readmission of patients within a short time of discharge. Careful analysis of
the key factors and health condition of the patient is required to understand
this situation. Based on the analysis, they prepared a dataset with patient
age, length of stay, admission acuity, comorbidity index score, gender, and
few other parameters based on insurance factors. The ultimate aim of this
task is to find the percentage of risk for readmission. Data mining and evo-
lutionary algorithms are applied to make a thorough analysis (Zheng et al.
2015). Hospital admission, stay duration, and readmission all go together to
encompass the complete healthcare of an individual. Various keywords that
relate to metaheuristic algorithms in healthcare include: chemotherapy, pro-
cessing, heuristic, admissions, analyze, compressing, environment, expecta-
tion, imaging, integration, mental, models, optimization, nature-inspired,
local search, bio-inspired, and so on. The list is very exhaustive and provides
an avenue for extensive research by way of the hybridization of various algo-
rithms. Home healthcare is a great leap by the healthcare industry to provide
immediate attention to a person in need. To provide the perfect healthcare in
a timely manner, with the available resources, is achievable with the help of
mathematical processes and metaheuristic s when fine-tuned as matheuristic
(Yeh et al., Lin 2007; Moussavi et al. 2019).
Home healthcare systems require more planning and accuracy in order to
meet the patient’s requirements in a more personal environment. Need to
work on with traversing from one place to another place both on the cost and
time perspective are the main ingredients to consider. Various factors like
planned distance, total distance, additional distance, and any sort of delay
make up the matheuristic model. Transportation cost is the primary criteria
for an organization to provide effective treatment. Along with this hybrid
genetic algorithm, combining the perspective of time and fuzzy concept
will enhance the system (Allaoua et al. 2013; Shi et al., 2017). It’s very certain
that metaheuristic algorithms will make the healthcare environment more
affordable and user friendly with faster diagnoses and better. Discussions
provided in this paper provide great support to the hybridization of meta-
heuristic algorithms and will glorify the research to a great extent.

3.7 Conclusion
Metaheuritics are the techniques directing the process of searching. Most of
the procedures follow a random process and therefore the result is a collection
36 Machine Learning for Healthcare

of random variables. When compared with techniques like optimization and

iteration methods, metaheuristics are able to provide satisfactory results with
less computational effort. Therefore, metaheuristics are the best approach for
task optimization. Much of the literature about metaheuristics is theoretical
in nature, explaining empirical results based on algorithm-based computer
experiments. But there are also some formal theoretical tests, frequently on
convergence and the potential to find the global optimum. Several metaheuris-
tic approaches have been published with statements of novelty and functional
effectiveness. In recent decades, the healthcare industry is facing a higher level
of uncertainty due to the appearance of new diseases and increases of casu-
alty. Analyzing the stored historic datasets help in predicting the future seri-
ousness with high levels of clarity. The evolution of metaheuristic algorithms
is interesting and innovative in nature and it takes care to revolve around
nature-inspired components. The healthcare industry along with Artificial
Intelligence concepts will reach great heights and the scope of research is vast.
Perfect analyzing of the attributes needing improvement will add a greater
impact on performance metrics and early disease diagnosis.

References
Abdelaziz, Ahmed, Mohamed Elhoseny, Ahmed S. Salama, and A.M. Riad. “A
machine learning model for improving healthcare services on cloud comput-
ing environment”. Measurement 119 (2018): 117–128.
Allaoua, Hanane, Sylvie Borne, Lucas Létocart, and Roberto Wolfler Calvo. “A
matheuristic approach for solving a home health care problem”. Electronic Notes
in Discrete Mathematics 41 (2013): 471–478.
Amirian, Pouria, Francois van Loggerenberg, Trudie Lang, Arthur Thomas, Rosanna
Peeling, Anahid Basiri, and Steven N. Goodman. “Using big data analyt-
ics to extract disease surveillance information from point of care diagnostic
machines”. Pervasive and Mobile Computing 42 (2017): 470–486.
Åström, Freddie, and Rasit Koker. “A parallel neural network approach to predic-
tion of Parkinson’s disease”. Expert Systems with Applications 38(10) (2011):
12470–12474.
Baro, Emilie, Samuel Degoul, Régis Beuscart, and Emmanuel Chazard. “Toward
a literature-driven definition of big data in healthcare”. BioMed Research
International 2015 (2015) 1–9.
Berger, Marc L., and Vitalii Doban. “Big data, advanced analytics and the future of
comparative effectiveness research”. Journal of Comparative Effectiveness Research
3(2) (2014): 167–176.
Carneiro, Allan C., José G.F. Lopes, Marcelo M.S. Souza, Jeová F. Rocha Neto, Flávio
H.D. Araújo, Romuere R.V. Silva, Fátima N.S. Medeiros, and Francisco N.
Bezerra. “Parameter optimization of a multiscale descriptor for shape analysis
on healthcare image datasets”. Pattern Recognition Letters 125 (2019).
Metaheuristic Algorithms  37

Chen, Jianguo, Kenli Li, Huigui Rong, Kashif Bilal, Nan Yang, and Keqin Li. “A dis-
ease diagnosis and treatment recommendation system based on big data min-
ing and cloud computing”. Information Sciences 435 (2018): 124–149.
Cosma, Georgina, David Brown, Matthew Archer, Masood Khan, and A. Graham
Pockley. “A survey on computational intelligence approaches for predictive
modeling in prostate cancer”. Expert Systems with Applications 70 (2017): 1–19.
Dai, Y., and G. Wang. “A deep inference learning framework for healthcare”. Pattern
Recognition Letters 22(9) (2018): 1–9.
de Ridder, Mischa, Alfons J.M. Balm, Ludi E. Smeele, Michel W.J.M. Wouters, and
Boukje A.C. van Dijk. “An epidemiological evaluation of salivary gland cancer
in the Netherlands (1989–2010)”. Cancer Epidemiology 39(1) (2015): 14–20.
Dokeroglu, Tansel, Ender Sevinc, Tayfun Kucukyilmaz, and Ahmet Cosar. “A sur-
vey on new generation metaheuristic algorithms”. Computers and Industrial
Engineering 137 (2019): 106040.
Galetsi, Panagiota, Korina Katsaliaki, and Sameer Kumar. “Big data analytics in
health sector: Theoretical framework, techniques and prospects”. International
Journal of Information Management 50 (2020): 206–216.
Hashem, Ibrahim Abaker Targio, Victor Chang, Nor Badrul Anuar, Kayode Adewole,
Ibrar Yaqoob, Abdullah Gani, Ejaz Ahmed, and Haruna Chiroma. “The role
of big data in smart city”. International Journal of Information Management 36(5)
(2016): 748–758.
Hegde, J., and B. Rokseth. “Applications of machine learning methods for engineer-
ing risk assessment – A review”. Safety Science 122 (2020): 104492.
https://www.vosviewer.com/ accessed January 13, 2020.
Juan, Angel A., Javier Faulin, Scott E. Grasman, Markus Rabe, and Gonçalo Figueira.
“A review of simheuristics: Extending metaheuristics to deal with stochastic
combinatorial optimization problems”. Operations Research Perspectives 2 (2015):
62–72.
Kouanou, Aurelle Tchagna, Daniel Tchiotsop, Romanic Kengne, Djoufack Tansaa
Zephirin, Ngo Mouelas Adele Armele, and René Tchinda. “An optimal big data
workflow for biomedical image analysis”. Informatics in Medicine Unlocked 11
(2018): 68–74.
Lee, June-Goo, Sanghoon Jun, Young-Won Cho, Hyunna Lee, Guk Bae Kim, Joon
Beom Seo, and Namkug Kim. “Deep learning in medical imaging: General
overview”. Korean Journal of Radiology 18(4) (2017): 570–584.
Liberatore, Matthew J., and Robert L. Nydick. “The analytic hierarchy process in
medical and health care decision making: A literature review”. European Journal
of Operational Research 189(1) (2008): 194–207.
Makwakwa, Lumbani, Mei-ling Sheu, Chen-Yuan Chiang, Shoei-Loong Lin, and
Peter W. Chang. “Patient and health system delays in the diagnosis and treat-
ment of new and retreatment pulmonary tuberculosis cases in Malawi”. BMC
Infectious Diseases 14(1) (2014): 132.
Moussavi, Seyed Esmaeil, M. Mahdjoub, and Olivier Grunder. “A matheuristic
approach to the integration of worker assignment and vehicle routing problems:
Application to home healthcare scheduling”. Expert Systems with Applications
125 (2019): 317–332.
Ngiam, Kee Yuan, and Wei Khor. “Big data and machine learning algorithms for
health-care delivery”. The Lancet Oncology 20(5) (2019): e262–e273.
38 Machine Learning for Healthcare

Nilashi, Mehrbakhsh, Othman Bin Ibrahim, Hossein Ahmadi, and Leila Shahmoradi.
“An analytical method for diseases prediction using machine learning tech-
niques”. Computers and Chemical Engineering 106 (2017): 212–223.
Nilashi, Mehrbakhsh, Othman Bin Ibrahim, Abbas Mardani, Ali Ahani, and Ahmad
Jusoh. “A soft computing approach for diabetes disease classification”. Health
Informatics Journal 24(4) (2018): 379–393.
Nilashi, Mehrbakhsh, Othman Ibrahim, Sarminah Samad, Hossein Ahmadi, Leila
Shahmoradi, and Elnaz Akbari. “An analytical method for measuring the
Parkinson’s disease progression: A case on a Parkinson’s telemonitoring data-
set”. Measurement 136 (2019): 545–557.
Palanisamy, Venketesh, and Ramkumar Thirunavukarasu. “Implications of big data
analytics in developing healthcare frameworks – A review”. Journal of King Saud
University-Computer and Information Sciences 31(4) (2019): 415–425.
Peeling, Rosanna W. “Diagnostics in a digital age: An opportunity to strengthen
health systems and improve health outcomes”. International Health 7(6) (2015):
384–389.
Pelteret, Jean-Paul, Bastian Walter, and Paul Steinmann. “Application of metaheuris-
tic algorithms to the identification of nonlinear magneto-viscoelastic constitu-
tive parameters”. Journal of Magnetism and Magnetic Materials 464 (2018): 116–131.
Rezk, Eman, Zainab Awan, Fahad Islam, Ali Jaoua, Somaya Al. Maadeed, Nan
Zhang, Gautam Das, and Nasir Rajpoot. “Conceptual data sampling for breast
cancer histology image classification”. Computers in Biology and Medicine 89
(2017): 59–67.
Shayanfar, Human, and Farhad Soleimanian Gharehchopogh. “Farmland fertility: A
new metaheuristic algorithm for solving continuous optimization problems”.
Applied Soft Computing 71 (2018): 728–746.
Shi, Yong, Toufik Boudouh, and Olivier Grunder. “A hybrid genetic algorithm for
a home health care routing problem with time window and fuzzy demand”.
Expert Systems with Applications 72 (2017): 160–176.
Tack, Christopher. “Artificial intelligence and machine learning| Applications in
musculoskeletal physiotherapy.” Musculoskeletal Science and Practice 39 (2019):
164–169.
Thai, L.H., T.S. Hai, and Nguyen Thanh Thuy. “Image classification using support
vector machine and artificial neural network”. International Journal of Information
Technology and Computer Science (IJITCS) 4(5) (2012): 32–38.
Tsai, Chun-Wei, Ming-Chao Chiang, Adlen Ksentini, and Min Chen. “Metaheuristic
algorithms for healthcare: Open issues and challenges”. Computers and Electrical
Engineering 53 (2016): 421–434.
Urdea, Mickey, Laura A. Penny, Stuart S. Olmsted, Maria Y. Giovanni, Peter Kaspar,
Andrew Shepherd, Penny Wilson et al. “Requirements for high impact diag-
nostics in the Developing World”. Nature 444(1s) (2006): 73.
Willems, Stefan M., Sanne Abeln, K. Anton Feenstra, Remco de Bree, Egge F. van
der Poel, Robert J. Baatenburg de Jong, Jaap Heringa, and Michiel W.M. van
den Brekel. “The potential use of big data in oncology”. Oral Oncology 98 (2019):
8–12.
Wills, Mary J. “Decisions through data: Analytics in healthcare”. Journal of Healthcare
Management 59(4) (2014): 254–262.
Wong, Zoie S.Y., Jiaqi Zhou, and Qingpeng Zhang. “Artificial intelligence for infec-
tious disease big data analytics”. Infection, Disease and Health 24(1) (2019): 44–48.
Metaheuristic Algorithms  39

Yeh, Jinn-Yi, and Wen-Shan Lin. “Using simulation technique and genetic algo-
rithm to improve the quality care of a hospital emergency department”. Expert
Systems with Applications 32(4) (2007): 1073–1083.
Zheng, Bichen, Jinghe Zhang, Sang Won Yoon, Sarah S. Lam, Mohammad
Khasawneh, and Srikanth Poranki. “Predictive modeling of hospital readmis-
sions using metaheuristics and data mining”. Expert Systems with Applications
42(20) (2015): 7110–7120.
4
Decision Support System to
Improve Patient Care

V. Diviya Prabha and R. Rathipriya

CONTENTS
4.1 Introduction................................................................................................... 41
4.2 Related Work.................................................................................................42
4.3 Feature Selection...........................................................................................42
4.3.1 Entropy Formula...............................................................................44
4.4 Experimental Setup...................................................................................... 47
4.5 Conclusion..................................................................................................... 47
References................................................................................................................ 51

4.1 Introduction
The amount of data is quickly increasing at a very fast pace in the healthcare
sector. The extraction of relevant data from high volume data is a challeng-
ing task. The novel physiognomies of medical data are challenging for data
mining. The wide variety and the huge volume of data is valuable only when
a useful pattern is extracted. The required model is available in the raw data
alongside data that is not useful. Obtaining such useful (Archenaa and Mary
Anita 2016; Malykh and Rudetskiy 2018) knowledge with a pre-existing data
mining approach that is crucial.
Similarly, forming decision-making tasks from raw data in various dimen-
sions of other data are essential. Moreover, there are numerous reports avail-
able in hospitals based across cities (Liu et al. 2018) and villages. Forming the
correct decision based on a patient’s data helps both patients and doctors get
a good result (Abraham 2016). High dimensional data with a large number of
features show us the importance of feature selection.
Over the last few decades, feature selection with machine learning
approaches an important area for research. The significance of choosing the
best features helps the doctors and patients understand the medical (Sasikala
et al. 2016) data in making the appropriate decision and diagnosis of infected
patients as soon as possible. There are several method filters, and wrapper
and embed methods for feature selection.

41
42 Machine Learning for Healthcare

This chapter discusses the likelihood of patients being readmitted to the

hospital after discharge, and how the knowledge of drugs taken by the
patient and medical data can help in making these predictions. Several pre-
vious studies motivate us. An entropy-based measure is useful with the inte-
gration of machine learning techniques.
It is analyzing data using attribute selections for a predictive model by
considering necessary attributes that support prediction and removing irrel-
evant attributes. Much of clinical data today consists of irrelevant attributes
that weaken the prediction level. Thus, the proposed approach concentrates
on decision entropy-based attribute selection. The subset of the attribute is
selected based on entropy value and given as an input to the machine learn-
ing algorithms such as logistic regression (Prabha et al. 2019), support vector
machines, decision trees, etc., for prediction of readmissions. This gives bet-
ter accuracy than the existing model. It also suits for increasing data dynami-
cally to handle new data for readmission prediction. This helps to reduce
readmission risk and improves patient care.
The chapter is organized as follows: In section two, the basic concepts of
entropy-based feature selection are explained. In section three, the algo-
rithms and flow of PySpark-based algorithms are evaluated, and concentra-
tions of features are reported. Section five concludes the paper.

4.2 Related Work
Frequent work has been carried out in the field of feature selection. The accu-
rate prediction model needs to identify the relevant feature (Xing et al. 2001).
The significant features represent the strengths and weaknesses of the fea-
tures. For subset feature selection, a DFL algorithm (Maji 2010) is used to find
the optimal features. Large datasets have more features, so it is important to
categorize the relevant features. Differentiation entropy (Cassara et al. 2016)
is used for feature subsets so that these important features can be selected.
The neighborhood entropy works better for classical game theory process. It
recommends that Shannon’s entropy works well only for nominal data, and
that it does not work well for other data values. A subset feature selection
using entropy (Zheng and Chee 2011) for a huge amount of data is the most
critical task here, and identifying the correct subset of feature is difficult in
this approach (Ahmed and Kumar 2018; Masoudi-Sobhanzadeh et al. 2019).

4.3 Feature Selection
The basic filter and wrapper methods are performed in prevision papers,
but that is not significant importance to feature selection. Feature selection
for high dimensional datasets plays an important role in disease prediction
Decision Support Systems  43

(Agrawal 2016; Li et al. 2016). In this paper, the main objective is to study the
different types of entropy in large datasets. Entropy is used to identify the
signs of attributes present in the dataset. The dataset is taken from the UCI
repository. The first important step in data mining is data preprocessing. The
main advantage is using Spark in Python is the PySpark. The first process is
to import PySparkin to the PySpark context, and creating the PySpark con-
text as SparkContext in the local environment.
The flowchart in Figure 4.1 represents the flow of the proposed method
in the PySpark environment. The datasets are preprocessed as pipeline
processes that combine multiple algorithms in the single process. Other
processes are also carried out: StringIndexer for character variables,
OnHotEncoder for the binary values and converting them to vectors as vec-
tor assembler. StringIndexer helps to convert the string values in the dataset
into numeric form; for example, gender features consists of male and female.
It converts this data into 0 for males and 1 for females. For OnHotEncoder,
VectorAssembler is used for preprocessing. All these processes are simulta-
neous processes that reduce the computation time of the model.

FIGURE 4.1
Steps for the proposed methodology.
44 Machine Learning for Healthcare

4.3.1 Entropy Formula
Entropy is an essential approach used to identify the uncertainty of the pre-
dicted variables. Here, decision entropy performs the same based on the tar-
get value. The measurement of entropy is used to select the best variable for
the target class. Different types of entropy are carried out to find the criteria
for best feature selection. Following this, for each variable, the entropy for-
mula is the following:
For each variable, the entropy is calculated as the following:
n

Entropy _ features = å - prob ( f ) log ( prob ( f ))

i =1
i 2 i

ïì1 H ( Entropy_features,target_variable ) > = 1

Decision_Entropy = í
ïî0 H ( Entropy_features,target_variable ) < 1

The entropy formula for decision entropy is formulated on the basis of if the
entropy_features based on the target variable. If the entropy_features value
is greater than 1 then the feature is selected, but if the entropy features based
on target variable are less than 1 then the feature value is set to 0 means
which the feature is rejected. The decision_entropy is used to identify rel-
evant features based on the target variable.
Figures 4.2–4.5 characterize the different types of entropy used to select
the best features for readmission prediction (Marcello and Battiti 2018).
These figures highlight each variable that supports the entropy method. Each

FIGURE 4.2
Shannon’s entropy.
Decision Support Systems  45

FIGURE 4.3
Relative entropy.

FIGURE 4.4
Boltzman’s entropy.

selected attribute promptly changes from one entropy method to another.

The cross-entropy method (Weiss and Dardick 2019) epitomizes the prob-
ability distribution of one method to other methods.
In the current work, different works based on entropy are obtained in the
proposed work decision entropy is computed to improve the predictive accu-
racy. The entropy values are calculated based on the target value. If the values
46 Machine Learning for Healthcare

FIGURE 4.5
Cross-entropy.

FIGURE 4.6
Decision entropy.

of the particular attribute are greater than 1 then that feature is selected for
prediction, otherwise the feature is rejected. The feature values are similarly
calculated. Figure 4.6 represents decision entropy, which provides a way to
identify the optimal features and improves the best features. The best 11
features are selected for the prediction of readmission.
Table 4.1 represents readmission predictions for hospitals using different
entropy methods. The proposed method of decision entropy is compared
Decision Support Systems  47

with the existing approach. The accuracy of the proposed approach is high
when compared with the existing approach. Shannon’s entropy, Boltzmann’s
entropy, and cross-entropy have a certain similar range of prediction.
Additionally, relative entropy and decision entropy have similar ranges, of
which overall performance accuracy of the proposed entropy performance is
better than the existing approach which is at 92% accuracy.

4.4 Experimental Setup
The readmission prediction is of significant importance for both hospitals
and patients. The goal of this chapter is to suggest a better prediction model
for understanding whether the patient is likely to be readmitted or not.
Table 4.2 indicates the different measures of the probability value of each
variable. From Table 4.2, it can be observed that each step of the variable
value consequently varies from one to another.
Table 4.3 represents the different measures signifying the importance of
the variable. Biased represents the difference between the predicted values
and target values. The bias value must be low for a good prediction model.
Here, most of the values of the variables are low. Similarly, other measures
such as co-variance, determination, t-tests, and p-values are calculated.
Figure 4.7 embodies the value of the density function used for each vari-
able. The graph represents each variable with the density function to rec-
ommend the importance of the function. Figure 4.8 represents the graph
plotting of the prediction of different entropy, which the proposed work per-
formance better compared to other work.

4.5 Conclusion
In this chapter, a novel approach of decision entropy–based logistic regres-
sion (DELR) is put forward as an appropriate method for readmission

TABLE 4.1
Accuracy Prediction Using Entropy Methods
Accuracy prediction (%)
Entropy Methods LG DT RF GBT Average Value
Shannon Entropy 66 67 63 67 65.75
Boltzmann Entropy 58 55 57 57 56.75
Cross Entropy 65 47 61 62 58.75
Relative Entropy 92 90 92 93 91.75
Decision Entropy 94 93 92 91 92.25
 48

TABLE 4.2
Concentration Measures for Variables
Concentration V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11
Herfindahl 1.503 0.73 1.81 6.22 2.77 6.67 0.0 0.01 2.088 4.62 3.58
N- Herfindahl 8.44 0.80 3.96 4.80 1.35 5.27 9.54 0.01 6.68 3.20 2.61
Rosenbluth 1.63 0.88 1.96 6.10 2.77 6.69 0.01 0.01 2.23 3.24 3.86
Gini 0.133 0.88 0.27 0.767 0.48 0.78 0.87 0.91 0.36 0.57 0.63
Ricci-Schutz 0.95 0.88 0.19 0.767 0.48 0.78 0.87 0.88 0.26 0.47 0.50
Atkinson 0.17 0.88 0.06 0.753 0.48 0.78 0.87 0.89 0.10 0.28 0.49
Kolm 54.38 0.03 9.84 0.756 0.120 0.06 0.44 0.08 1.88 2.05 0.80
Co-efficient of variation 0.24 2.83 0.52 1.83 0.97 1.92 2.59 3.42 0.68 1.50 1.23
Squared Coefficient 0.05 8.02 0.27 3.38 0.95 3.71 6.72 11.63 0.47 2.25 1.52
Machine Learning for Healthcare
TABLE 4.3
Measures for Variables
Decision Support Systems 

Title V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11

Biased Variance 2.6 0.09 6.6 0.24 0.24 0.1 0.36 0.3 8.64 2.3 3.07
Biased SD 15.97 0.31 8.28 0.49 0.49 0.3 0.60 0.60 2.93 5.22 1.75
Co variance 0.56 0.00 0.15 0.00 0.00 0 0.04 0.0 0.08 0.22 0.02
Correlation 0.07 0.00 0.03 0.03 0.03 0 0.14 0.1 0.06 0.08 0.02
Determination 0.005 7.25 0.00 0.00 0.00 0.0 0.02 0.0 0.03 7.98 0
T- test 19.09 2.26 10.2 9.76 9.13 2.6 38.8 38.8 16.3 2.37 6.52
p-value 4.27 0.02 1.60 1.57 6.81 0.0 0 0 4.71 0.01 6.72

Correction made in indicated in red text are correct and can be updated.
49
50 Machine Learning for Healthcare

FIGURE 4.7
Value of data in the form of density.

FIGURE 4.8
Accuracy Prediction.
Decision Support Systems  51

prediction of patients. This method is compared with the existing method.

The overall performances are better; as it is capable of rejecting irrelevant
features and selecting only relevant features. The DELR method’s perfor-
mance has good accuracy of prediction at 92% effectiveness. Selecting better
features improves the classification accuracy.

References
Agrawal, Rashmi. “A modified K-nearest neighbor algorithm using feature optimi-
zation”. International Journal of Engineering and Technology 8(1) (2016): 28–37.
Archenaa, J., and E. A. Mary Anita. “Interactive big data management in healthcare
using spark”. In: Proceedings of the 3rd International Symposium on Big Data and
Cloud Computing Challenges (ISBCC–16’). Springer, Cham, 2016.
Cassara, Pietro, Alessandro Rozza, and Mirco Nanni “A cross-entropy-based method
to perform information-based feature selection.” arXiv Preprint ArXiv:1607.07186
(2016).
Frandsen, Abraham Jacob. “Machine learning for disease prediction”. Theses and
Dissertations, Brigham Young University, 5975 (2016).
Kauser Ahmed, P., and N. Senthil Kumar. “A comprehensive review of nature –
inspired algorithm for feature selection”. In: Handbook of Research on Modeling,
Analysis, and Application of Nature-Inspired Metaheuristic Algorithms, Sujata Dash,
B. K Tripathy (Eds). IGI Global, 2018.
Li, Falcao, Zan Zhang, and Chenxia Jin. “Feature selection with partition differ-
entiation entropy for large-scale data sets”. Information Sciences 329 (2016):
690–700.
Liu, Kanghuai, et al. “Big medical data decision-making intelligent system exploiting
fuzzy inference logic for prostate cancer in developing countries”. IEEE Access
7 (2018): 2348–2363.
Maji, Pradipta. “Mutual information-based supervised attribute clustering for
microarray sample classification”. IEEE Transactions on Knowledge and Data
Engineering 24(1) (2010): 127–140.
Malykh, V. L., and S. V. Rudetskiy. “Approaches to medical decision-making based
on big clinical data”. Journal of Healthcare Engineering 2018 (2018): 6313–6322.
Marcello, Andrea, and Roberto Battiti. “Feature selection based on the neighbor-
hood entropy”. IEEE Transactions on Neural Networks and Learning Systems 29(12)
(2018): 6313–6322.
Masoudi-Sobhanzadeh, Yosef, Habib Motieghader, and Ali Masoudi-Nejad.
“FeatureSelect: A software for feature selection based on machine learning
approaches”. BMC Bioinformatics 20(1) (2019): 170.
Prabha, V., Diviya, and R. Rathipriya. “Readmission prediction using hybrid logis-
tic regression”. In: International Conference on Innovative Data Communication
Technologies and Application. Springer, Cham, 2019.
Sasikala, S., S. Appavu Alias Balamurugan, and S. Geetha. “Multi Filtration Feature
Selection (MFFS) to improve discriminatory ability in clinical data set”. Applied
Computing and Informatics 12(2) (2016): 117–127.
52 Machine Learning for Healthcare

Weiss, Brandi A., and William Dardick. “An entropy-based measure for assessing
fuzziness in logistic regression”. Educational and Psychological Measurement 76(6)
(2016): 986–1004.
Weiss, Brandi A., and William Dardick. “Separating the odds: Thresholds for entropy
in logistic regression”. The Journal of Experimental Education 76(6) (2019): 1–22.
Xing, E., M. Jordan, and R. Karp. “Feature selection for high-dimensional genomic
microarray data”. In: Proceedings of the 18th International Conference on Machine
Learning. Morgan Kaufmann Publishers Inc., Williamstown, MA, 28 June – 1
July 2001, pp. 601–608.
Zheng, Yun, and Chee Keong Kwoh. “A feature subset selection method based on
high-dimensional mutual information”. Entropy 13(4) (2011): 860–901.
5
Effects of Cell Phone Usage on Human
Health and Specifically on the Brain

Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi,

Mehmood Naqvi and Shakeel Ahmed

CONTENTS
5.1 Introduction................................................................................................... 53
5.2 Background.................................................................................................... 55
5.3 Radiation Produced by a Mobile Phone.................................................... 56
5.4 MATLAB Tools.............................................................................................. 57
5.4.1 Problem Statement............................................................................ 57
5.4.2 Research Objective............................................................................ 57
5.5 State-of-the-Art Research and Technology............................................... 58
5.6 Discussion of Tools....................................................................................... 59
5.7 Methodology................................................................................................. 59
5.7.1 Quantitative Approach.................................................................... 60
5.7.2 Design Research................................................................................ 60
5.8 Method of Data Collection.......................................................................... 60
5.8.1 Sampling Technique......................................................................... 60
5.8.2 Sample Size........................................................................................ 60
5.8.3 Instrument for Data Collection....................................................... 60
5.8.4 Research Model................................................................................. 60
5.9 K-Means Clustering...................................................................................... 61
5.10 Result and Discussion..................................................................................64
5.11 Conclusion.....................................................................................................64
References................................................................................................................65

5.1 Introduction
Software engineering is about the theories, methods, procedures, processes,
and tools used for the development of software. Software development as a
method of software engineering goes hand-in-hand with research and devel-
opment. Digital image handling is an inconceivable, broad, and limitless field
which can utilize different applications which incorporate everything from

53
54 Machine Learning for Healthcare

the detection of fingerprint identification, to brain cancer detection (Norman

et al. 2018; Das et al. 2017; Shree and Kumar 2018).
Malignancy is depicted as a startling tissue growth. Cerebral malignant
growths are unusual masses of tissue where cells aggregate. What’s more,
they multiply in an unpredictable way – one that is obviously not under the
control that ordinary cells are subjected to. The cell can be assessed as each
organism’s main basic unit. The human body contains about 100 billion cells
and each has its own individual capabilities that relate to the body’s ideal
functioning, and these cells need to split in a measured way to form health
new cells. In addition, each cell is highly fragmented and even terminated
from time-to-time in order to create new cells. This process sometimes leads
to the development a square of unnecessary cancer tissue. Cancer can hap-
pen anywhere in the body. Brain cancer for good and life can be assumed
to be a true cancer. In terms of their starting lands and risks as cancers are
shown (Lee et al. 2018; Klank et al. 2011).
The extent of radio-recurrence vitality relies on development of initial
growth of cancer in terms of degree and type of utilization, of mobile phone,
which is connected to PDA towers, from client to RF wave base station.
Physicists have isolated each possible joint cause between versatile radiation
and natural systems, for example., physical twists of proteins, inception of
hailing pathways, and docking with receptors on cell layers. For each case,
the discovered nature of portable radiation has been insignificant in affect-
ing any dangerous change (Abdullah et al. 2013). A few examinations have
looked into the effect of radio-recurrence fields on cerebral electrical work,
psychological capacity, rest, pulse, circulatory strain, and more. This work
was completed at the National Institute of Technology (NIT), Kerala, India. In
this study, brain pain, dizziness, hip depth, and chest weight were assessed
among the “constant” mobile phone users. An examination during the test
details the increase of the heartbeat when holding a mobile phone near the
chest and near the head. A huge difference was not found in the opposite
conditions without a mobile phone. (Klank et al. 2011). Investigations into
participant health history did not show any consistent evidence of adverse
health effects from exposure to radio frequency fields, other than the effect
of heating the tissues. In addition, researchers were unable to provide proof
that links the final conditions and outcomes to electromagnetic field expo-
sure and any declared side effects, for example, electromagnetic hypersensi-
tivity among mobile users. (Kinaci et al. 2018).
Epidemiological research is looking at the risk of long-term exposure to
radio frequencies. For the most part, it has sought to link brain tumors to
the use of mobile phones. However, the same amount of cancer remains
undetectable for many years after exposure to the cause of the tumor. Since
mobile phones were not commonly used until the mid-1990s, epidemiolog-
ical studies have been limited to cancers originating in recent times. The
results of animal studies have shown that there is no risk of cancer due to
long-term exposure to radio frequency fields (Lee et al. 2018). A few trials
Cell Phone Usage Effects on Health  55

have attempted to find other possible long-term health effects of exposure

to mobile radiation. In a test conducted in Amritsar, Punjab, India, the link
between the use of mobile phones (exposure to RF radiation) and DNA chro-
mosomes damage the lymphocytes of mobile phone customers. This risk
may have long-term consequences for increasing tumor risk or other age-
related changes. (Ferlay et al. 2015).

5.2 Background
Therapeutic images provide important information about the human heart.
Mobile phones operate by transmitting signals and allowing the use of radio
frequency waves by people from nearby base stations. This is actually an
exposure of electromagnetic radiation, which can come from radio waves
and microwaves and more. Exposure related to non-ionizing radiation can
come from radio waves, microwaves, clear, soft, and warm waves, and RF
waves. These types of radiation simply do not give off enough radiation
to cause the development of a strong DNA lesion within the muscles. The
strength of radio frequency waves differ in the strongest (ionizing) radiation-
related forms (such as x-rays, gamma light, and bright radiation [UV]) that
can dilute the binding substances in DNA. The real reason for this research
is how microwaves work. Nevertheless, the amount of radiation released by
mobile phones is much lower and inadequate to greatly alter body tempera-
tures. Mobile phone use has increased dramatically (Kinaci et al. 2018; Ferlay
et al. 2015; Gelb et al. 2018; Kant et al. 2018). The increasing use of this technol-
ogy was a cause for concern in relation to wellbeing and safety. Earlier than
expected in the 1990s, low-level exposure to an electromagnetic radio (RF) field
has become of serious consequence on wellbeing. This was primarily exam-
ined by a group of specialists and an evaluation was conducted for possible
aggressive effects on the wellbeing of multi-use communications. Radiation
from every source is a critical secretion. X-rays are a perfect sample of radia-
tion, but sunlight and heat in our bodies are always progressing strongly.
There is a difference in radiation from “high-vital” high-energy radiation to
“low-vital” low-energy radiation. This includes the proposed method, where
electromagnetic radiation and gamma radiation are classified as being very
vital and are programs with enough vitality that you can shed a good elec-
tron from your (ionized) molecule. This can easily damage DNA within cells,
which can simply recognize progress. RF radiation is located at the lower end
of impedance using the electromagnetic obtained, this a type of non-ionizing
radiation (Abdullah et al. 2012; Miller 2016). Non-ionizing radiation provides
enough basic elements to move or make vibrations of particles inside a mol-
ecule, yet they are insufficient for ionization (forcing particles to force them,
for example, electrons). RF radiation has a higher centralization than is seen
56 Machine Learning for Healthcare

in the use of a noticeable degree of low-emission electromagnetic radiation.

The basic elements are less than found in some impeccable types of non-ion-
izing radiation in the same cycle, because of a light that can be recognized
alongside infrared radiation. Ionizing radiation generally has more energy.
With the possibility of RF radiation being absorbed into a giant mammals’
aggregate of substances containing mineral water, for example, reinforce-
ments, fluids, and body tissues, then heat will be generated to create abnor-
mality after burning the normal cells of the brain. This can easily provoke
explosions along with tissue pain. Despite the fact that the radiation of the RF
won’t cause harm to the DNA in cells due to the manner in which ionizing
radiation happens, there has been concern that the numerous ways of non-
ionizing radiation may have more of a general impact that can cause disease
in a couple of cases. The International Agency for Research on Cancer (IARC)
renewed this cause authenticity, which was normal, which could be conceiv-
able and outlines a general investigation of the connection between PDAS
use notwithstanding the danger of mind (malignant) growth. Thereafter, the
intervention became a general method of case-control studies that focus on
the types of paradoxical carcinomas in tissues, which can be revealed by tons
of centralized radio-frequencies caused by phones: brain carcinomas (glioma
and meningioma), acoustic nerve (oral tumor), and parotid appendix. The
aim was to understand whether the use of cells increases the risk of these
cancers as well as whether the remote-controlled RF power will be cancerous
(Jian et al. 2018; van der Kleij et al. 2018; Gamage and Ranathunga 2017).

5.3 Radiation Produced by a Mobile Phone

Unlike televisions, warning systems, PCs, and all other electrical contrap-
tions, cell phones (also called mobile phones) are electronic devices that
generate electromagnetic radiation using radio frequency (RF) imperatives.
We work at low intensity (under 1 watt) by transmitting and tolerating elec-
tromagnetic radiation at the end of the range of the radiofrequency (RF).
It is understood that radiation which is assigned to be “ionizing” can be
devoured by isolated tissue and split molecules, such as gamma pillars and
x-bars, and can cause growth. The tension is that the PDA and its array con-
traction (the radiation wellspring) are held close to the head. The damage to
the DNA particles is accepted as the explanation. The radiation that a cell
phone uses is part of the comparable electromagnetic range, and isn’t ion-
izing. In this manner, the US FDA can control these contraptions to ensure
that the radiation doesn’t get to a level of threat to customers, but just once
the nearness of a general prosperity danger has been developed. RF essen-
tialness was incorrectly believed to equivalently cause development (Gupta
and Pahuja 2017; Heinen et al. 2016).
Cell Phone Usage Effects on Health  57

5.4 MATLAB Tools
MATLAB is a multi-viewpoint numerical processing framework and type
of prohibitive lingo programming made by Math Works. MATLAB awards
coordinate controls, cutoff point plotting and information, use of computa-
tions, enhancement of UIs, and interfacing with activities written in different
vernaculars, including: C, C++, C #, Java, Fortran, and Python (Lavanyadevi
et al. 2017). Regardless of how normal MATLAB is for numerical process-
ing, an optional device compartment uses the MuPAD significant motor,
enabling access to the enrolling limits of the agent. An additional package
for dynamic and embedded systems, Simulink, incorporates graphic multi-
space reenactment and model-based arrangement (Soobia et al. 2019a,b).

5.4.1 Problem Statement
Cancer is characterized as an unexpected tissue change. Brain cancer is an
irregular mass of tissue in which cells begin dividing in an uncontrollable
manner, evidently without the normal influences that govern normal cells.
The cellular can be described as every organism’s primary unit. The human
body carries about a hundred billion cells, each with its own very precise
skills in relation to the frame’s most necessary functions, and those cells
require fissures to cut up in a measured way to viably form new cells. In
addition, each cell is highly fragmented and even terminated from time-to-
time in order to create new cells. This process sometimes leads to the devel-
opment a square of unnecessary cancer tissue.
The issue of mobile phone usage for more than 50 minutes, brain tissues on
a single side of the head the mobile phone, radio wire processed more glu-
cose when contrasted with the tissues about the backward side of the brain.
In this research, the causes of brain cancer disease due and their links to the
mobile phone – in relation to this expansion in glucose metabolism – are still
unknown. The type of brain cancer rapidly created in the brain due to the
usage of cellular phone is also unknown, and this is one of the major motives
of our research area. The author is using a silicon-based neuron cell chip for
the purpose of storing the damage cancer cells, which are created by cel-
lular phones. It is being investigated how this relates to the growth of brain
cancer. Further, authors have also used this chip to investigate the activity
of cancerous cells and how we uncover the disadvantage of damaged cells
of brain cancer.

5.4.2 Research Objective

• To find the brain cancer with the help of ROI algorithms.

• To evaluate the effectiveness of image testing in brain-imaging tech-
nologies that utilize MATLAB tools and data mining.
58 Machine Learning for Healthcare

5.5 State-of-the-Art Research and Technology

There are several steps of cancer, which can occur by using a mobile phone
from beginning stage to final stage of cancer due to extreme use of mobile
phone in terms of RF waves. We also realize that the most common types
in adults are meningiomas that are considered glioma called glioblastoma
in several ways. Many types are extremely rare. Cancers of the brain may
occur at any age. For example, vaccines are often more usual in adolescents
with unit arthroplasty, and some are also necessary in adults. In general,
age-related adult cancers are more common. Diffuse brain cancers and
dangerous main brain cancers are unfortunately common than favorable
(Li et al. 2015; Abdullah et al. 2013; Bishop and Favaro 2011). Early signs of
brain cancer can include migraine sensations. This is due to the increase
in intracranial mass. These indicators may come and go at the beginning
of cancer growth, and in the morning they tend to be at their worst. Piracy
can exacerbate pain during breathing and crouching. Additionally, epilep-
tic seizures are often exacerbated or triggered. As the cancer progresses to
a large mass, lethargy and “laziness” may occur. When the cancer grows,
the surrounding brain tissue is likely to be affected (Saeed and Jafri 2015;
Soobia et al. 2019). The separate parts of the cerebrum control the capaci-
ties of various parts of the body. From that point forward, depending on
the impact on the part of the brain and the extent of that area’s overall
impact, the symptoms change on a case-to-case basis. For instance, symp-
toms often include: muscle shortening in the arm, leg, or part of the face
or eyes, issues with removal, joint grouping, vision, hearing, voice, cor-
respondence, or drinking. Further symptoms may include: loss of smell,
unsteadiness, change in mood, disarray or shortcoming of memory, numb-
ness in parts of the body, feelings of perplexity, changes in character, and
indications of hormonal changes in pituitary disease (Usman and Rajpoot
2017; 21Lee et al. 2017; Oliveira et al. 2018). This is can be examined by moni-
toring of brain capabilities, neurological developments, reactions, vision,
etc. MRI scans or CT scans are regular tests that are carried out to confirm
or prevent brain cancer to examine more microscopic elements (Saeed and
Jafri 2015; Soobia et al. 2019). If the cancer is distinguished, more scans and
tests can be done. In some rare cases, to gather more data about cancer,
a PET scan or angiography may be conducted. It may be assumed that a
biopsy would confirm the type of cancer. A biopsy is a procedure where
a tiny tissue sample is removed from a body part (Pim et al. 2012; Soobia
et al. 2012; Liu et al. 2018). Under a microscope the sample is examine for
odd cells. A small surgery, usually with sedatives, is required to obtain a
brain biopsy. A small area of the skull is reduced to allow a fine needle to
be inserted in order to withdraw a small tissue sample. By observing the
acquired cells of a biopsy, a final diagnosis can be made about if there is a
Cell Phone Usage Effects on Health  59

threat, and if so, an identification of the stage and type of cancer (Liu et al.
2016; Liang et al. 2018).
Information processing in the central nervous system interferes primarily
with the interconnected relationship between the neurons. This relationship
distinguishes how well large-scale social events of neurons can regulate,
respect, and implement complex sensory boundaries, including learning and
memory. The synaptic system between any particular groups of neurons isn’t
hard-wired; instead it is comprised of an abnormal state of flexibility, which
constitutes the neuroplasticity in humans that gives us the ability to learn and
memorize. While there is extensive research into the beginning of cellular
and semi-nuclear commencement of synaptic versatility – at the level of junc-
tions of neurons or smaller structures – the examination of larger neuronal
groups has been demonstrated in reality tests. The ability to assess the activi-
ties of large neural frames at the same time and in a non-prominent way is
a fundamental basis for understanding how neural frameworks work at the
system level. Here, we present the leaps forward in the range of many bionic
mixtures that have been properly linked to neural frames with silicon devices
to detect the performance of synaptic-related neurons (Hazra et al. 2017).

5.6 Discussion of Tools
Generally, the tools discussed so far do not scale very well and there is no
official standard of comparison available, thus, this is one of our objectives
in this chapter. However, as per the literature studies regarding cancer
investigation methods, where the MATLAB tools/code for the previously
mentioned tools are concerned, there is very little information available on
recommended resources. Tool comparison as used in different techniques is
the most difficult task, because each investigation/detection technique has
a different method. Apart from that, each tool has its own way of investiga-
tion/detection, so focusing on specific tasks is quite a difficult task.

5.7 Methodology
The research is conducted by first gathering the patient’s cancer-related
data. The data collected includes attributes like patient’s gender, age, stage
of cancer, and demography, etc. An initial bibliographic research approach
can be implemented alongside the study of literature. In the previous stages,
research methods and qualitative designs were used.
60 Machine Learning for Healthcare

5.7.1 Quantitative Approach
For a quantitative approach I intend to use the following methodologies.
Investigation or detection of damage cells of cancer:

1. Examine the brain cell through the usage of MATLAB software

tools.
2. Propose a benchmark comparing the damaged and undamaged
cancer cells with different tools.
3. Analyze the available data mining techniques used by available
tools.

5.7.2 Design Research
Design research is sometimes called “enhancement research”. It emphasizes
problem solving. In our domain it is intended to create a detection or inves-
tigation process.

5.8 Method of Data Collection

This is secondary data that is taken from different medical organizations of
Malaysia.

5.8.1 Sampling Technique
The sampling technique used to study sampling is random sampling and
participants are randomly taken from various hospitals in Malaysia.

5.8.2 Sample Size
For data collection, the sample size is 150. The target population for select-
ing the sample was cancer patients and they were transferred to hospitals in
various medical institutions and hospitals in Malaysia.

5.8.3 Instrument for Data Collection

The software SSPS Statistics is used as the instrument for the data collection.

5.8.4 Research Model
The assembly is used as a research model. Aggregation is a very powerful
data mining technique that is used to define similarities between different
Cell Phone Usage Effects on Health  61

groups. There are several aggregation algorithms in which the k-means algo-
rithm is used to define hidden patterns of data. In my research, the k-mean
algorithm explores invisible information by taking an attribute such as gen-
der, age, cancer stage, demography, etc. This model identifies cancer at a very
early stage.

5.9 K-Means Clustering
The K-Means aggregation algorithm was created in 1976 by MacQueen. It
is an unsupervised collection that performs a certain number of separate
groups and (non-dynamic) levels. The strategy is based on a clear and basic
way to handle the request of a specific dataset through a specific number of
groups (k-mean) previously created. K-mean estimates identify objects with-
out k-objects, going to the initial meeting center. The accompanying step is
to take each point that has a place within a specific set of data and move it to
the closest center in perspective, near the article to the meeting center, using
Euclidean division. After passing each item, it is necessary to recalculate the
positions of the new group. The strategy is repeated until there is no match
in the beam centers’ k. This estimate reduces an objective limit known as the
permissible limit for the square screw that was determined by the transition
(Soobia et al. 2019).
c ci
J (v ) = åå( x - v )
i j =0
i j

Where,
‘||xi – vj||’ is the Euclidean distance between xi and vj,
here are the main steps for the K-Means algorithm:

1. Dataset distribution in Group K.

2. For each data point in the dataset:
• Measure the distance from each group to the data point;
• If the set is near the data point, and if you leave it, the data point
will move to the nearest set.
3. Perform the previous step until any data point moves from one
group to another. This stage shows that it is the end of the assembly
process and that all groups are stable.
4. The choice of the primary section can greatly affect the resulting
final groups, in the form of consistency and distances within the
group and between groups as shown in Table 5.1:
TABLE 5.1
62

Statistical Data of Independent Sample

Independent Samples Test
Levene’s Test for
Equality of Variances T-test for Equality of Means
95% Confidence Interval
of the Difference
Sig. Mean Std. Error
F Sig. t Df (2-tailed) Difference Difference Lower Upper
NART Assumption of Eq. 80.640 .000 7.483 42 .000 .727 .097 .531 .923
Variance
without Assumption 7.483 21.000 .000 .727 .097 .525 .929
of Eq. Variance
RAVLT Assumption of Eq. 13.644 .001 1.763 42 .085 .227 .129 −.033 .487
Variance
without Assumption 1.763 37.978 .086 .227 .129 −.034 .488
of Eq. Variance
Trail A Assumption of Eq. 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Variance
without Assumption 2.898 33.600 .007 .364 .125 .109 .619
of Eq. Variance
Trail B Assumption of Eq. 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Variance
without Assumption 2.898 33.600 .007 .364 .125 .109 .619
of Eq. Variance
(Continued)
Machine Learning for Healthcare
TABLE 5.1 (CONTINUED)
Statistical Data of Independent Sample
Independent Samples Test
Levene’s Test for
Equality of Variances T-test for Equality of Means
95% Confidence Interval
of the Difference
Sig. Mean Std. Error
F Sig. t Df (2-tailed) Difference Difference Lower Upper

HPTR9 Assumption of Eq. 23.579 .000 2.748 42 .009 .364 .132 .097 .631
Variance
without Assumption 2.748 37.192 .009 .364 .132 .096 .632
Cell Phone Usage Effects on Health 

of Eq. Variance
HPTL9 Assumption of Eq. 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Variance
without Assumption 2.898 33.600 .007 .364 .125 .109 .619
of Eq. Variance
COWAT Assumption of Eq. 8.090 .007 1.600 42 .117 .227 .142 −.059 .514
Variance
without Assumption 1.600 40.810 .117 .227 .142 −.060 .514
of Eq. Variance
WDRT Assumption of Eq. 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Variance
without Assumption 2.898 33.600 .007 .364 .125 .109 .619
of Eq. Variance
mwalk10 Assumption of Eq. 22.120 .000 2.411 42 .020 .318 .132 .052 .584
Variance
without Assumption 2.411 37.277 .021 .318 .132 .051 .585
of Eq. Variance
63
64 Machine Learning for Healthcare

5.10 Result and Discussion

Data mining has an unusual centralization of the scope of the drug, and
addresses a comprehensive methodology that requires a deep understand-
ing of the needs of restorative departments. The learning obtained through
the use of data mining techniques can be used to determine productive
decisions that improve compliance with human administration affiliation
and patient prosperity. Data mining requires appropriate development and
research techniques, as well as systems for reporting and viewing, after
which results can be measured. Data mining, once started, handles an end-
less cycle of learning dissemination. As for affiliations, it offers one of the
main things that make the business strategy not so bad. Today, there have
been several attempts aimed at the applicable use of data mining in social
protection of societies. The main ability of this framework lies in the possibil-
ity of examining a persuasive case in datasets in the social protection space.
These illustrations can be used for clinical research. In any case, open and
useful data is widely distributed; it is different and huge in nature. This data
should be collected and stored in data warehouses in the formed structures,
and it can be provided with a specific final goal for forming the specialist
office information system. The progress of the data mining process is a pro-
cedure that the researcher organizes in an attempt to find new and hidden
cases in the data, from which they can learn. This is a discovery that can help
provide patients with restorative and distinctive healthcare organizations.

5.11 Conclusion
This type of study portrays cancer levels of serotonin using impression seg-
mentation of brain cancer among MRI images, and illustrates the potential
impacts of the proposed solution. In addition, taking up studies in view of
security issues and substantial unique obstacles is an enormous task. The
area of this study has always been the classification of brain cancer tactics
from side-to-side. The MRI data from the Brain Web Database provides a
valid example of the representation of brain cancer of the MRI. In this spe-
cific research work, we handle basic brain images that are thought of as can-
cers, choosing cancer section area that may relate to mobile phone use. These
sorts of images are, on a very basic level, in correspondence with the sort
and size of cancer enormous and should decide how progressive the cancer
is by detecting or identifying cancer with the usage of the neuron silicon
chip–based technology (using the neuron network). In this research, we have
focused on how far reaching cancer is and we identify the range of tumor
area from initial to final stage for particular dimensions of cancer cell. By
Cell Phone Usage Effects on Health  65

utilizing a new technology such as neuron networks, we have tested mul-

tiple types of skulls to detect brain cancer location. In brain cancers, there
are needed to include the main area regarding. Our research’s accuracy is the
most important measure of accomplishment. Thus, this study proports that
MRI scans gather the best images and outcomes. In addition, it is important
to store the sample or patches of brain cancer that identify how to control the
brain cancer or reduce the dangers of cellular technology.

References
Abdullah, Afnizanfaizal, et al. “An improved swarm optimization for parameter
estimation and biological model selection”. PLoS One 8(4) (2013): 4.
Abdullah, Afnizanfaizal, et al. “Cerebrospinal fluid pulsatile segmentation-a
review.” In: The 5th 2012 Biomedical Engineering International Conference. IEEE,
2012.
Bishop, Tom E., and Paolo Favaro “The light field camera: Extended depth of field,
aliasing, and superresolution”. IEEE Transactions on Pattern Analysis and Machine
Intelligence 34(5) (2011): 972–986.
Chaudhary, P., and R. Agrawal “A comparative study of linear and non-linear clas-
sifiers in sensory motor imagery based brain computer interface”. Journal of
Computational and Theoretical Nanoscience 16(12) (2019): 5134–5139.
Chaudhary, P., and R. Agrawal “Emerging threats to security and privacy in brain
computer interface”. International Journal of Advanced Studies of Scientific Research
3 (2018): 12.
Das, Suman, et al. “Detection and area calculation of brain tumour from MRI images
using MATLAB”. International Journal 4 (2017): 1.
Ferlay, Jacques, et al. “Cancer incidence and mortality worldwide: Sources, methods
and major patterns in GLOBOCAN 2012”. International Journal of Cancer 136(5)
(2015): E359–E386.
Gamage, P.T., and Dr Lochandaka Ranathunga “Identification of brain tumor using
image processing techniques”. Faculty of Information Technology, University of
Moratuwa, 2017. https://www. researchgate. net/publication/276133543.
Gelb, Sivan, et al. “Mechanisms of neuropsychiatric lupus: The relative roles of
the blood-cerebrospinal fluid barrier versus blood-brain barrier”. Journal of
Autoimmunity 91 (2018): 34–44.
Gupta, Anjali, and Gunjan Pahuja. “Hybrid clustering and boundary value refine-
ment for tumor segmentation using brain MRI”. In: IOP Conference Series:
Materials Science and Engineering, Vol. 225, No. 1. IOP Publishing, China, 2017.
Hazra, Animesh, et al. “Brain tumor detection based on segmentation using
MATLAB.” In: 2017 International Conference on Energy, Communication, Data
Analytics and Soft Computing (ICECDS). IEEE, India, 2017.
Heinen, Rutger, et al. “Robustness of automated methods for brain volume measure-
ments across different MRI field strengths”. PLoS One 11(10) (2016): 10.
Jian, Wen-xuan, et al. “Potential roles of brain barrier dysfunctions in the early stage
of Alzheimer’s disease”. Brain Research Bulletin 142 (2018): 360–367.
66 Machine Learning for Healthcare

Kant, Shawn, et al. “Choroid plexus genes for CSF production and brain homeostasis
are altered in Alzheimer’s disease”. Fluids and Barriers of the CNS 15(1) (2018): 34.
Khan, A. Raouf, Noor Zaman, and Saira Muzafar “Health hazards linked to using
mobile cellular phones”. Journal of Information and Communication Technology 2(2)
(2008): 101–108.
Khotanlou, Hassan, et al. “3D brain tumor segmentation in MRI using fuzzy clas-
sification, symmetry analysis and spatially constrained deformable models”.
Fuzzy Sets and Systems 160(10) (2009): 1457–1473.
Kinaci, Ahmet, et al. “Effectiveness of dural sealants in prevention of cerebrospinal
fluid leakage after craniotomy: A systematic review”. World Neurosurgery 118
(2018): 368–376.
Klank, R.L. et al. Migration in confinement: A micro-channel-based assay. In: Wells,
C.M., M. Parsons, editors. Cell Migration: Developmental Methods and Protocols.
Humana Press: Totowa, NJ, pp. 415–434, 2011.
Lavanyadevi, R., et al. “Brain tumor classification and segmentation in MRI images
using PNN.” In: 2017 IEEE International Conference on Electrical, Instrumentation
and Communication Engineering (ICEICE). IEEE, Yogyakarta, Indonesia 2017.
Lee, Chi-Hoon, et al. “Segmenting brain tumors with conditional random fields and
support vector machines.” In: International Workshop on Computer Vision for
Biomedical Image Applications. Springer: Berlin, Heidelberg, China, 2005, 1–12.
Lee, Su Yeon, et al. “Regulation of tumor progression by programmed necrosis”.
Oxidative Medicine and Cellular Longevity 2018 (2018).
Li, Guodong, et al. “Automatic liver segmentation based on shape constraints and
deformable graph cut in CT images”. IEEE Transactions on Image Processing: A
Publication of the IEEE Signal Processing Society 24(12) (2015): 5315–5329.
Liang, Fan, et al. “Abdominal, multi-organ, auto-contouring method for online adap-
tive magnetic resonance guided radiotherapy: An intelligent, multi-level fusion
approach”. Artificial Intelligence in Medicine 90 (2018): 34–41.
Liu, Jia, et al. “A cascaded deep convolutional neural network for joint segmenta-
tion and genotype prediction of brainstem gliomas”. IEEE Transactions on Bio-
Medical Engineering 65(9) (2018): 1943–1952.
Mendrik, Adriënne M., et al. “MRBrainS challenge: Online evaluation framework
for brain image segmentation in 3T MRI scans”. Computational Intelligence and
Neuroscience, Japan, (2015): Japan, 1–12.
Miller, Kimberly. Cancer Treatment & Survivorship Facts & Figures. American Cancer
Society Estimated Numbers of Cancer Survivors Report, (2016): 1–44.
Moeskops, Pim, et al. “Automatic segmentation of MR brain images with a convo-
lutional neural network”. IEEE Transactions on Medical Imaging 35(5) (2016):
1252–1261.
Oliveira, Gustavo Casagrande, Renato Varoto, and Alberto Cliquet Jr. “Brain tumor
segmentation in magnetic resonance images using genetic algorithm cluster-
ing and adaboost classifier”. BIOIMAGING. 2018.
Saeed, Soobia, and Afnizanfaizal Abdullah “Recognition of brain cancer and cere-
brospinal fluid due to the usage of different MRI image by utilizing support
vector machine”. Bulletin of Electrical Engineering and Informatics 9(2) (2020):
619–625.
Saeed, Soobia, Afnizanfaizal Abdullah, and N.Z. Jhanjhi “Analysis of the lung can-
cer patient’s for data mining tool”. IJCSNS 19(7) (2019a): 90.
Cell Phone Usage Effects on Health  67

Saeed, Soobia, and Afnizanfaizal Bin Abdullah “Investigation of a brain cancer

with interfacing of 3-dimensional image processing.” In: 2019 International
Conference on Information Science and Communication Technology (ICISCT). IEEE,
Pakistan, 2019.
Saeed, Soobia, Afnizanfaizal Abdullah, and N.Z. Jhanjhi “Implementation of Fourier
transformation with brain cancer and CSF images”. Indian Journal of Science and
Technology 12(37) (2019b): 37.
Saeed, Soobia, and Raza Jafri “Estimation of brain tumor using latest technology of
mobile phone”. Journal of Information and Communication Technology (JICT) 9(1)
(2015): 32–09.
Saunders, Norman R., et al. “Physiology and molecular biology of barrier mecha-
nisms in the fetal and neonatal brain”. The Journal of Physiology 596(23) (2018):
5723–5756.
Sharma, Geeta, et al. “A detailed study of EEG based brain computer interface.”
In: Proceedings of the First International Conference on Information Technology and
Knowledge Management, New Dehli, India, Vol. 14, Dehli, India, 2017.
Shree, N. Varuna, and T.N.R. Kumar “Identification and classification of brain tumor
MRI images with feature extraction using DWT and probabilistic neural net-
work”. Brain Informatics 5(1) (2018): 23–30.
Usman, Khalid, and Kashif Rajpoot “Brain tumor classification from multi-modality
MRI using wavelets and machine learning”. Pattern Analysis and Applications
20(3) (2017): 871–881.
van der Kleij, Lisa A., et al. “Fast CSF MRI for brain segmentation; Cross-validation
by comparison with 3D T1-based brain segmentation methods”. PLoS One 13(4)
(2018): 4.
6
Feature Extraction and Bio Signals

A. Mary Judith, S Baghavathi Priya, N. Kanya, and Jyotir Moy Chatterjee

CONTENTS
6.1 Introduction................................................................................................... 69
6.2 Feature Extraction......................................................................................... 71
6.2.1 Common Spatial Patterns................................................................ 72
6.2.2 Adaptive Common Spatial Patterns............................................... 72
6.2.3 Adaptive CSP Patches ..................................................................... 72
6.2.4 Canonical Correlation Analysis ..................................................... 73
6.2.5 Band Power Features ....................................................................... 73
6.2.6 Adaptive Band Power Features....................................................... 74
6.2.7 Time Point Features ......................................................................... 74
6.2.8 Time Points with Adaptive XDAWN............................................. 74
6.3 Feature Selection and its Approaches........................................................ 75
6.3.1 Filter Approach ................................................................................ 76
6.3.2 Wrapper Approach........................................................................... 76
6.4 Conclusion..................................................................................................... 78
References................................................................................................................ 78

6.1 Introduction
To use a Brain-Computer Interface (BCI), two stages required are: 1) an offline
preparing stage organization during which the arrangement is balanced and
2) an online stage where the structure sees brain development examples and
makes an understanding of them for the activity of the PC. The practical
online BCI structure begins with the patient conveying a specific example,
which is followed by evaluating the EEG signals. EEG signals are routinely
pre-taken care of using diverse, and spatial channels (Blankertz et al. 2008),
alongside the component expulsion from the signs. Before making an inter-
pretation of the sign in order to provide an input on whether a particular
mental action has been performed or not, the EEG signal features are first
arranged (Lotte et al. 2007) by a classifier.
Spatial channels are occasionally used for preprocessing the signal fea-
tures so as to find a change to make a lot of bogus channels by direct mix of

69
70 Machine Learning for Healthcare

the underlying channels. These channels improve the sign to clamor propor-
tion (SNR) and decrease the data dimensionality. Spatial channels depend
upon subject-unequivocal information, with the objective that different
strategies were made to achieve a particular degree of client freedom. The
ghastly information depicts how the force changes and is used in some spe-
cific recurrence groups. The transient records depict how the important EEG
signals differ with time. This demonstrates the signal utilization at selective
time factors or in particular time frames
The essential step which includes extraction targets depicts the sign
through several relevant qualities. Such features should lead to the informa-
tion being embedded in the sign, which is imperative when depicting the
mental exercises to recognize, rejecting the antiques, and other non-critical
information. Features expelled are usually distributed into a component
vector. Features can likewise be tuned by distinguishing the most note-
worthy channels or recurrence groups to indicate unmistakable cerebrum
movement.
A learning calculation is stood up to with the goal of picking the capabili-
ties and to concentrate on the component determination issue. The regulated
learning calculations have a standard plan to improve the classification of
exactness on example test information. Rather than trying to achieve exact-
ness, we prioritized finding the capabilities that are most relevant and used
them for learning. In administered machine learning, an enlistment calcula-
tion is normally provided with fixed preparing of an informational index, in
which each datum is depicted by the vector esteems and the features’ class
name. For example, in clinical visualization issues, the capabilities would
potentially comprise of the stature, weight, and so forth, and the class mark
may demonstrate whether the healthcare professional diagnosed the disor-
der of the individual.
As it is understandable to acquire better accuracy for different features,
the component subsets need not be one of a kind. For an ideal element subset
created by a list of capabilities, choice calculation delivers the most advanced
conceivable precision. Alongside the most effective capacity, the pertinence
of highlight is likewise a crucial imperative. A component is exceptionally
applicable if its disposal brings about execution disintegration. Additionally,
an element has less importance in the event that it isn’t relevant in every
case and in the presence of capabilities which are worse than the general
execution. On the off chance that the element is firmly important, it can’t be
expelled without exactness misfortune. Powerless pertinence infers that the
trademark can infrequently add to forecasting exactness.
Different exceptional BCI classifiers (Nicolas-Alonso and Gomez-Gil 2012)
exist and are being used. Numerous methodologies often utilize the linear
discriminant analysis (LDA) as a classifier. The creators (Shenoy et al. 2006;
McFarland et al. 2011) demonstrated that the LDA classifier is retrained on
unmistakable informational indexes which deliver a result with a decent
classification execution. Another avenue is to utilize a lot of covariance
Feature Extraction and Applications of Bio Signals 71

networks as clarified in Vidaurre et al. (2011a and b). The support vector
machine (SVM) (Li et al. 2008) and other probabilistic neural classifiers
(Millan and Mouriño 2003) are intermittently used to for updates. There are
other classification calculations, for example, clump calculations are used
to hold the whole dataset and in the event that new records are brought to
the preparation dataset, at that point the classification model is recomputed.
Additionally, an online classifier intends to explicitly refresh the classifica-
tion model as new records are received.
Noticeably, in order to improve spatial channels, which have become a key
section of classifications, these readings are joined in order to demonstrate
these estimations, to report how they were used for BCIs and identify the
outcomes. This section is aware of their points of interest and disadvantages
and it aims to give examples of how and when to use a specific classification
procedure as well as highlighting some of the challenges that must be over-
come to enable further advancement in the sign classification. Tripathy et al.
(2019) endeavored to progress over explores concerning machine learning for
enormous data informative and different techniques with respect to intro-
duce day figuring circumstances for various social applications. Chatterjee
(2018a) gave a short review of how machine learning can be utilized in bioin-
formatics. Chatterjee (2018a) endeavored to give a reasonable and thorough
understanding of the IoT in BD structure, examining its different issues
and challenges, and concentrating on giving potential recommendations of
machine learning procedures.
This section is composed as follows: segment two introduces the ordinar-
ily utilized EEG highlight extraction and its strategies, segment three briefs
on the determination systems, and, finally, the procedure is outlined.

6.2 Feature Extraction
In BCI pattern recognition and machine learning systems utilize a classifier,
but the additionally incorporate extraction and determination procedures to
show the signs in a smaller and more appropriate manner. In particular, the
EEG signals are consistently isolated under time region (band-pass channel)
and spatial area (spatial channel) before the capabilities are expelled from
the consequent signs. The element choice calculation perceives and chooses
the best capabilities which, at that point, prepares the classifier. This fea-
tures different techniques for highlighting extraction strategies to pick the
most material features and ways to evaluate the resulting pattern recogni-
tion. Despite the fact that there are various component extraction strategies
available, the two most regular types of features that could work well are
time point features. Additionally, frequency band power features could be
an option.
72 Machine Learning for Healthcare

6.2.1 Common Spatial Patterns

Common spatial patterns (CSP) are utilized (Ramoser et al. 2000) to separate
features from EEG pointers. The features obtained are absolutely based at a
definitive part. Partition resources of CSP are generally attractive for sepa-
rating two populaces of EEG. The CSP and spatial channels learn the prepa-
ration realities rapidly and don’t perform appropriately with an enormous
amount of heterogeneous measurements (Congedo 2013). A more prominent
and pertinent approach is to promptly regularize the CSP trademark and not
the covariance grids (Lotte and Guan 2011). Traditional unaided procedures
which understand PCA or managed systems that are understanding of CSP
can consequently be utilized. Dornhege et al. (2004) have proposed an invari-
ant CSP (iCSP), that regularizes the basic spatial patterns’ objective compo-
nent and decreases the impact of old rarities. The regularization approach is
currently proving to be progressive and successful.

6.2.2 Adaptive Common Spatial Patterns

Adaptive common spatial patterns are compelling for patients with hear-
ing difficulty and for crippled patients. An adaptive CSP (ACSP) system was
used to manage the intra and bury changeability issues in the bio signal. In
order to improve the segregation power, the adaptive CSP makes an analysis
of the objective subject and simultaneously overhauls the spatial channels. It
additionally supports in characterizing the single preliminary EEG records
under testing conditions while preparing data as accessible from one sub-
ject and with no preparation records from the objective subjects. The ordi-
nary CSP and its multisubject expansions assess the spatial channels and
are additionally arranged for classification (Devlaminck et al. 2011). During
classification, in order to improve the multisubject by large execution of CSP,
the ACSP approach joins the objective subjects’ data with the CSP learning
on preliminary premise.

6.2.3 Adaptive CSP Patches

Adaptive CSP patches refer to the capability of the CSP analysis to make
small arrangements of channels (patches) and the blending of features. The
theory is that applying CSP examination on just two or three channels, the
threat of overfitting is diminished with basic CSP, which is then resolved
on every single open channel. This is in light of the fact that the number
of parameters to fit for each fix isn’t as much as that for CSP. The patches
can join a substitute number of encompassing channels. Moreover, the focal
points of the patches can be picked dependent on the number of channels
open and in the assignment. For each fix, different channels identical to the
number of included channels are a delayed consequence of the CSP assess-
ment. From those channels, one for each class is picked – for instance, two
Feature Extraction and Applications of Bio Signals 73

channels for each fix are achieved. From the ensuing gathering of channels,
of the six most helpful ones, three for each class are chosen. Vidaurre et al.
(2011) also examined the co-adaptive preparing (both machine and user are
continually learning) by using adaptive features and an adaptive LDA classi-
fier. This enabled a couple of users, who were previously unfit to control the
BCI, to become able to do so. Co-adaptive preparing that utilizes adaptive
CSP patches has been show to be extensively and progressively beneficial
(Sannelli et al. 2011). In particular, CSP patches can be seen as a CSP regular-
ize without the need of assessing hyper parameters. It uses covariance grids
of lower estimation, improving the gauge of the parameters and thus, basi-
cally improving the customer’s introduction using the preparation data. It
might be assumed that the CSP approach is useful with very few channels or
with multi-channel accounts but it is astoundingly proficient, as the amount
of alteration data required is thoroughly diminished – interestingly with the
bleeding edge.

6.2.4 Canonical Correlation Analysis

Also known as CCA, canonical correction analysis is used for a spatial chan-
nel. The goal of CCA is to identify direct changes which builds an understand-
ing of the association between crude EEG signal and the perfect waveform to
get a successful spatial channel. The standard CCA is summarized to tensor
CCA and multiset CCA which are adequately used to amass the SSVEP flags
in BCI (Zhang et al. 2017). Tensor canonical correlation analysis (TCCA), and
its successor multiset-standard canonical correlation analysis (MsetCCA), is
one of the profitable systems utilized for recurrence recognition in signals.
The MsetCCA procedure displays various linear changes which complete
the spatial sifting to extend the correlation among canonical variates. This
isolates the SSVEP essential capabilities from different arrangements of EEG
data gained at a similar boost recurrence. What’s more, the improved ref-
erence signals are molded by the mix of regular features and furthermore
depend absolutely on those features. Expansive test analysis with signals
indicate that the tensor CCA and MsetCCA system improves the precision
of SSVEP recurrence recognition in correlation with the standard canonical
technique, especially with few channels and little available time. The perva-
sive results exhibited by the tensor CCA and MsetCCA strategy are a prom-
ising chance for precision in SSVEP repeat recognition (Zhang et al. 2014).

6.2.5 Band Power Features

Band power features refers to the signs power for a particular repeat band
in a given channel, which is clear up at the midpoint or over a specific time-
frame. These features can be prepared in various ways and are extensively
used for BCIs’ oscillatory development, for instance changes in EEG signal
74 Machine Learning for Healthcare

amplitudes. With such a limit, the band power features are the best quality
level fof eatures for BCI subjects’ mental imagery, where some BCI focuses
on mental states by disentangling the mental job that need to be done using
developments or emotions, or for steady-state visually evoked potential
(SSVEP) signals. Various ways to deal with producing band power features
from EEG signals are clarified in Brodu et al. (2011). Regardless, at first the
EEG signals are band-pass separated from a particular channel to a recur-
rence band which then squares the ensuing sign to process the signal force
lastly midpoints over some stretch of time (for example 1s).

6.2.6 Adaptive Band Power Features

Adaptive band power features highlight and hold up a non-control state,
taking up the least amount of time, requiring no BCI ace, and having online
capacities subject to only two terminals. The classifiers work commendably
in the independently directed globally. The structure performs preliminary
based exception rejection and readies a linear discriminant assessment clas-
sifier subject to an auto-picked logarithmic band-power. The BCI at first iso-
lated a total of six logarithmic band-powers included in the gatherings from
9 to 13 and 16 to 26 Hz (Faller et al. 2012) from bipolar acceptances at C3 (FC3
– CP3), Cz (FCz – CPz), and C4 (FC4 – CP4). The system proceeded to pick the
most extraordinary discriminability as shown by the Fisher model.

6.2.7 Time Point Features

Time point features are a connection of EEG tests from all channels.
Ordinarily, these capabilities are expelled after the pre-arrangement, promi-
nently down-testing and band-pass or low-pass filtration. These features
mastermind the Event Related Potential that changes the EEG signals ampli-
tude time at an occasion (Lotte 2014). P300 spellers utilizes these features
most as often as possible.

6.2.8 Time Points with Adaptive XDAWN

The purpose of xDAWN is to find a change that updates the isolation among
signals and antiquities and licenses to diminish the data. The channels
could be found by extending the signal-to-noise notwithstanding noise pro-
portion (SSNR) as talked about by the summarized Rayleigh technique. In
the standard xDAWN figuring (Rivet et al. 2009), this procedure is tended
to by uniting the QR matrix decomposition (QRD) and the singular value
decomposition (SVD). The event-related potentials are regularly found by
the xDAWN model. The following are recognized as its downsides.
Right off the bat, counts are done in bunch model that stores all data in
memory. Furthermore, if the signal fluctuates or additional data should
be joined to ad lib the channel, the channels must be recomputed. This is
Feature Extraction and Applications of Bio Signals 75

normally not common sense in light of the essential computational effort

it requires. Regardless, the solidification of additional information is indis-
pensable for adjustment. As opposed to using the QRD and SVD to find the
perfect channels, a consistent methodology that relies upon the recursive
least squares (RLS) technique (Rao and Principe 2001) can be used. This strat-
egy has the going with focal points conversely with the customary method:
as required for modification, it grants to combine new data into the chan-
nel estimation framework and also, it consolidates the refreshed coefficients
that changes with the advancing states. The gradual estimation lessens
the amount of required memory needed for the computation of the ideal
channels.
The adaptive xDAWN (axDAWN) widens the xDAWN spatial channel
using a recursive least square approach and is viewed as a novel spatial
channel. This upgrade has huge points of interest where it allows the
slow estimation of the channel coefficients which are profitable for the
memory and computational exertion, and empowers the solidification
of new data to alter the present channel to the swaying conditions or
another subject.

6.3 Feature Selection and its Approaches

This section displays the ordinarily used component determination systems,
as these features are normally the commitment to classifiers. The element
choice advances are applied after the extraction to pick the capabilities with
various advantages.

1. At first, among the various features that can be isolated from EEG
signals, few may be excessive or distinctive to the mental states
highlighted by the BCI.
2. Additionally, the number of parameters that will be improved by
the classifier is identified with the number of features. Lessening the
number of features prompts the classifier to propel the least param-
eters. Thus, this declines possible overtraining effects and therefore
improves the execution – especially there are fewer planning tests.
3. Thirdly, from a data extraction point of view, if only a few features
are picked and situated, it is less difficult to identify which highlight
has high relevance to the objective cerebrum state.
4. Fourthly, a model with fewer features and less parameters conveys
brisk distinguishing proof for another example, as it should be com-
pelling by calculation.
5. Finally, amassing and memory of data will be decreased.
76 Machine Learning for Healthcare

6.3.1 Filter Approach
Channel strategies rely upon proportions of association between the capa-
bilities and the objective independent of the classifier used. The confirmation
coefficient, which is processed by squaring the Pearson correlation coeffi-
cient, is used as an element situating rule (Hastie et al. 2001). The coefficient
of confirmation could likely be used for a two-class issue, stamping −1 or
+1 as two distinct classes. The linear conditions among features and target
classes are perceived by the correlation coefficient. The clear non-linear pre-
processing technique could be applied to misuse non-linear associations. In
a similar manner, situating criteria subject to information speculation can be
used, for instance on the normal information between each component and
the objective (Peng et al. 2005). Many channels highlight choice strategies
and request estimation of likelihood densities and the joint thickness of the
element and class name of the data. The main course of action is to speak to
the features and target class names. The additional aim is to inexact their
densities with a nonparametric technique, for instance, Parzen windows. If
the densities are evaluated by an ordinary appropriation and the result is
achieved through common information, it would look like being obtained by
the coefficient of correlation. Channel approaches have linear and multifac-
eted natures concerning the amount of capabilities. Regardless, this would
provoke a decision on repetitive capabilities.
The essential inconvenience of this methodology is the place it completely
neglects the effects of the chosen highlight sets of the acceptance calculations’
execution. Strategies that fall under the channel approach are: 1) The FOCUS
estimation, initially described for without noise, are Boolean spaces which
completely see all arrangements of features, picking the most immaterial list
of capabilities which are satisfactory to distinguish the mark of value on all the
preparation set; and 2) The relief estimation, which apportions a pertinence
weight to each list of capabilities that are expected to demonstrate the central-
ity of the component to the objective information. This technique follows a
randomized figuring. It plays out an irregular test on the preparation set and
the pertinence value is refreshed. refreshes the significance, regards subject to
the difference between the picked event, and the two nearest instances of the
proportionate and converse class. On the off chance that the greatest features
are relevant to the idea, it would choose most of them – notwithstanding the
reality where just a few are needed for basic idea clarification.

6.3.2 Wrapper Approach
Wrapper approaches unravel to the detriment of a progressively drawn out
estimation time. A classifier is used this way to deal with recover include
sets. The wrapper system chooses a lot of features, moves them to the clas-
sifier for preparation, and checks the consequent exhibition. Finally, it drops
the pursuit or proposes another subset if the standard isn’t satisfied.
Feature Extraction and Applications of Bio Signals 77

An enlistment calculation of the wrapper approach exists within the ele-

ment choice calculation. This calculation looks through the element subset
using the acceptance calculation to break down and evaluate the capabilities.
The wrapper strategy is a direct methodology where the enlistment estima-
tion is taken as a black box. To complete a preparation with different courses
of action of features ousted from the data, the acceptance count is performed
on the information. The element subset with the most significant evaluation
is picked as the end and puts forward which acceptance count will be per-
formed. The consequent classifier is additionally evaluated on an unused
autonomous test set.
This methodology prompts a hunt of the potential contentions which
require a space express, a hidden express, an end state, and a web searcher
(Kohavi and John 1997). The chosen hunt space affiliation makes each state
indicate a list of capabilities. All of the listed capabilities in a state deter-
mine if the component is accessible (1) or missing (0). The accessibility of a
state would be either resolve, incorporate, or eradicate a list of capabilities
from the state consequently clinging to the most frequently used hunt space
in measurements. The target of this hunt is to find the state with the most
raised appraisal, using a heuristic way to deal with it directly. The exactness
estimation is utilized as both the heuristic and the appraisal work, as the
actuated classifiers’ real precision isn’t known.
The heuristic capacity makes the classification exactness cross-approve the
smaller datasets as often as possible, and more often than the big datasets.
Since smaller datasets take less effort to prepare, the time taken to assess
the exactness (the consequence of the acceptance count time and the cross-
endorsement span) doesn’t turn out to be exorbitantly fast. This heuristic tech-
nique has a disadvantage where smaller datasets experience cross-approval
to overcome the greatest distinction producing the least information. For
greater datasets, the wrapper approach can be utilized to save significant
extra time. Likewise, the embedded method fuses the component choice and
the appraisal in a particular system.
The univariate includes determination estimation and surveys the par-
ticular force of each element autonomously. By then, the best individual
element is chosen. Univariate methodologies are uncommonly speedy and
computationally compelling, yet they are simultaneously risky. Since only
individual element proficiency is considered, the potential redundancies or
complementarities among the features are ignored. The multivariate compu-
tations evaluate the various classes of features and hold the supreme features.
This calculation utilizes proportions of classification on the preparation set
or multivariate shared information measures for the count of the subset of
features. This exhibition measure engages to truly consider the impact of
redundancies or complementarities among features. Evaluating the value
of subsets of features prompts incredibly high computational necessities.
Multivariate methodologies normally rely upon heuristics or game plans to
diminish the number of subsets to evaluate. They are currently hazardous,
78 Machine Learning for Healthcare

yet generally speaking garner a lot of ideal exhibitions over univariate strate-
gies. In any case, the multivariate methodologies may be too deferred for use
when the number of features is high.

6.4 Conclusion
Filtering sought after by band force and time focuses include extraction are
the most notable extractions used in the EEG-dependent on bio signal han-
dling in comparison to the other segment types that have been examined and
used. The most generally utilized extraction is network features. This ele-
ment type gauges the association and synchronization among signals across
different sensors and recurrence bands, which could be assessed using stage
lock values and direct exchange capacities. It has been distinguished that a
blend of various features, for example, time focuses and band force, or band
force and network features, and so on, bring about the most accurate clas-
sification in correlation of utilization to a solitary component type. Merging
different element types extends dimensionality; consequently, it requires the
assurance of the most significant features to keep away from dimensionality
issues. Systems to diminish the dimensionality issues are additionally delin-
eated right now. Choice of features have given huge upgrades to BCI, as clari-
fied in section three. For example, the LDA for BCI of P300 framework and
recurrence assurance following the sifting strategies. This section likewise
discusses the element choice strategies which will be done following the ele-
ment extraction process. Among all the determination systems examined,
the wrapper methodologies (for example, the channel choice dependent on
SVM, extraction by linear regressor highlight choice by hereditary calcu-
lations and P300, or highlight choice by developmental calculations under
multiresolution analysis) are the generally utilized strategies. Unmistakably,
metaheuristic strategies are becoming progressively more likely to be con-
ceivably used for highlight assurance to keep up a vital good way from the
curse-of-dimensionality.

References
B. Blankertz, R. Tomioka, S. Lemm, M. Kawanabe, and K.-R. Muller. Optimizing spa-
tial filters for robust EEG single-trial analysis. IEEE Signal Processing Magazine,
25(1), 41–56, 2008.
N. Brodu, F. Lotte, and A. Lécuyer. Comparative study of band-power extraction
techniques for motor imagery classification. In: Computational Intelligence,
Cognitive Algorithms, Mind, and Brain (CCMB) IEEE Symposium on, 1–6. IEEE,
Paris, France, 2011.
Feature Extraction and Applications of Bio Signals 79

J. Chatterjee. IoT with big data framework using machine learning approach.
International Journal of Machine Learning and Networked Collaborative Engineering,
2(02), 75–85, 2018a.
J. M. Chatterjee. Bioinformatics using machine learning. Global Journal of Internet
Interventions and IT Fusion, 1(1), 28–35 2018b. [ISSN: 2582-1385 (online)].
M. Congedo. EEG Source Analysis. Habilitation `a Diriger des Recherches (HDR). Univ.
Grenoble Alpes, Grenoble, France, TEL, 2013.
D. Devlaminck, B. Wyns, M. Grosse-Wentrup, G. Otte, and P. Santens. Multisubject
learning for common spatial patterns in motor-imagery BCI. Computational
Intelliegnce and Neuroscience, 2011(217987), 1–9, 2011.
G. Dornhege, B. Blankertz, G. Curio, and K. R. Muller. Boosting bit rates in nonin-
vasive EEG single-trial classifications by feature combination and multi-class
paradigms. IEEE Transactions on Bio-Medial Engineering, 51(6), 993–1002, 2004.
J. Faller, C. Vidaurre, T. Solis-Escalante, C. Neuper, and R. Scherer. Autocalibration
and recurrent adaptation: Towards a plug and play online ERD-BCI. IEEE
Transactions on Neural Systems and Rehabilitation Engineering 20(3), 313–319, 2012.
T. Hastie, R. Tibshirani, and J. Friedman. The Elements of Statistical Learning. Springer,
New York, New York, 2001.
R. Kohavi and G. H. John. Wrappers for feature subset selection. Artificial Intelligence,
97(1), 273–324, 1997. Relevance.
Y. Li, C. Guan, H. Li, and Z. Chin. A self-training semi-supervised SVM algorithm and
its application in an EEG-based brain computer interface speller system. Pattern
Recognition Letters, 29(9), 1285–1294, 2008.
F. Lotte. A tutorial on EEG signal-processing techniques for mental-state recognition
in brain– computer interfaces. In: Guide to Brain-Computer Music Interfacing,
133–161. Springer, 2014.
F. Lotte, M. Congedo, A. Lécuyer, F. Lamarche, and B. Arnaldi. A review of classi-
fication algorithms for EEG-based brain-computer interfaces. Journal of Neural
Engineering, 4(2), R1–R13, 2007.
F. Lotte and C. Guan. Regularizing common spatial patterns to improve BCI designs:
Unified theory and new algorithms. IEEE Transactions on Bio-Medical Engineering,
58(2), 355–362, 2011.
D. J. McFarland, W. A. Sarnacki, and J. R. Wolpaw. Should the parameters of a BCI
translation algorithm be continually adapted? Journal of Neuroscience Methods,
199(1), 2011.
J. R., Millan and J. Mouriño. Asynchronous BCI and local neural classifiers: An over-
view of the adaptive brain interface project. IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 11(2), 159–161, 2003.
L. F. Nicolas-Alonso and J. Gomez-Gil. Brain computer interfaces, a review. Sensors,
12(2), 1211–1279, 2012.
H. Ramoser, J. Muller-Gerking, and G. Pfurtscheller. Optimal spatial filtering of single
trial EEG during imagined hand movement. IEEE Transactions on Rehabilitation
Engineering, 8(4), 441–446, 2000.
Y. Rao and J. Principe. An RLS type algorithm for generalized eigen- decomposition
in neural networks for signal processing XI, 2001. In: Proceedings of the 2001 IEEE
Signal Processing Society Workshop, North Falmouth, MA, 263–272, 2001.
B. Rivet, A. Souloumiac, V. Attina, and G. Gibert. xDAWN algorithm to enhance evoked
potentials: Application to brain computer interface. Biomedical Engineering, IEEE
Transactions On, 56(8), 2035–2043, 2009.
80 Machine Learning for Healthcare

H. Peng, F. Long, and C. Ding. Feature selection based on mutual information: Criteria
of max- dependency, max-relevance, and min-redundancy. IEEE Transactions on
Pattern Analysis and Machine Intelligence, 27(8), 1226–1238, 2005.
C. Sannelli, C. Vidaurre, K.-R. Mu üller, and B. Blankertz. CSP patches: An ensemble
of optimized spatial filters. an evaluation studies. Journal of Neural Engineering,
8, 2011.
P. Shenoy, M.Krauledat, B.Blankertz, R. P.Rao, and K. Müller To-wards adaptive clas-
sification for BCI. Journal of Neural Engineering, 3(1), 2006.
H. K. Tripathy, B. R. Acharya, R. Kumar, and J. M. Chatterjee. Machine learning on big
data: A developmental approach on societal applications. In: Big Data Processing
Using Spark in Cloud, 143–165. Springer, Singapore, 2019.
C. Vidaurre, M. Kawanabe, P. von Bünau, B. Blankertz, K. R. Müller. Toward unsuper-
vised adaptation of LDA for brain computer interfaces. Biomedical Engineering,
IEEE Transactions On, 58(3), 587–597, 2011a.
C. Vidaurre, C. Sannelli, K.-R. Müller, and B. Blankertz. Co-adaptive calibration to
improve BCI efficiency. Journal of Neural Engineering, 8(2), 025009, 2011b.
Y. Zhang, G. Zhou, J. Jin, X. Wang, and A. Cichocki. Frequency recognition in SSVEP-
based BCI using multiset canonical correlation analysis. International Journal of
Neural Systems, 24(04):1450013, 2014.
Y. Zhang, G. Zhou, J. Jin, Y. Zhang, X. Wang, and A. Cichocki. Sparse Bayesian multi-
way canonical correlation analysis for EEG pattern recognition. Neurocomputing,
225, 103–110, 2017.
7
Comparison Analysis of Multidimensional
Segmentation Using Medical
Health-Care Information

Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi,

Memood Naqvi, and Azeem Khan

CONTENTS
7.1 Introduction................................................................................................... 81
7.2 Literature Review.........................................................................................83
7.2.1 Static Structure of Literature Review with Another
Research Comparison......................................................................84
7.3 Methodology.................................................................................................84
7.3.1 Original Result of Image Testing in Binary Transformation......... 86
7.3.2 High Dimension Structured Graphs.............................................. 87
7.3.2.1 Grab-Cut.............................................................................. 87
7.4 Algorithm....................................................................................................... 87
7.5 Result Comparison and Discussion........................................................... 88
7.6 Conclusion..................................................................................................... 89
Acknowledgments................................................................................................. 91
References................................................................................................................ 91

7.1 Introduction
Graph cutting option in Image Segmented program, graph cut is a semi-
automatic segmentation technology that can be used by a researcher to
separate an image into front and back components. We can draw lines in
the image, called scribbles, to determine what is placed in the foreground
and what is placed in the background. The segmented image automatically
divides based on the scribble and displays the split image. The researcher
can improve segmentation by drawing more scribbles on the image until
they are satisfied with the result.
Graph cut technology applies graphics theory to image processing to
achieve rapid segmentation. The technique creates a graphic for the image

81
82 Machine Learning for Healthcare

where each pixel is a loop connected to a weighted edge. The more tightly
the pixel is bound the more weight. The algorithm cuts the weak edges, split-
ting the objects in the image. The split image uses a specific set of the graph
cut algorithm called “slow adjust”. The information for graphics segmenta-
tion on segmentation technology, such as graphics segmentation “grab-cut”,
is an image selected area.
The integrated graphics clipping algorithms have been successfully applied
to a wide range of vision and graphics problems. This article focuses on the
simplest graphic cutting app: segmenting an object in image data. In addition
to its simplicity, this application embodies the best features of harmonic graph-
ics cutting methods of vision: the optimum global level, practical efficiency,
and numerical durability, the ability to integrate a wide range of signals, opti-
cal restrictions, untied topological character sectors, and the applicability of
ND problems. It has also been proven that the graphics-based methods used
to extract objects have exciting links with previous hash methods, such as
snakes, active geodesic systems, and level groups. Improved division ener-
gies with realistic fragments consolidate limit association and locale-based
properties similarly as Mumford-Shah’s utilitarian style. We give the motiva-
tion and a point-by-point specialized portrayal of the fundamental consonant
advancement system for picture division by cutting s/t illustrations.
Figure 7.1 shows the process of graph cutting and also the selected region
of cutting area as we can use this tool for selecting the specific area for detect-
ing the disease.
The high-dimensional segmentation process is used to cut graphics and
find image quality. All images use the same process but the quality is dif-
ferent across four dimensions. After implementing the images, we can find

FIGURE 7.1
Graph cutting using MATLAB.
Comparison Analysis of Multidimensional Segmentation 83

something unique to choose the pictures Or, the quality of all the images
improved compared to the 3D images, but if we use the color images in 4D
segmentation, the result is more better then black and white images espe-
cially The quality of the images varies in colon versus white and black, as
mentioned in the experimental results.

7.2 Literature Review
Segmentation is one of the most significant assignments in the field of PC
vision and has been studied for a long time. One of the best-known image
splitting techniques is grab-cut (Saeed et al.), which is a moderate way to split
the foreground and background of 2D images. This is already implemented in
many photo editing software applications. Grab-cut depends on graphic cut-
ting algorithms (Boykov and Jolly 2001; Boykov and Kolmogorov 2004; Boykov,
Veksler, and Zabih 2001). Graphics cuts can be applied to dimensional infor-
mation, including pictures, video successions, and 3D structures (Gamage
and Ranathunga 2017), and can likewise be stretched out to different ticks
(Mendrik et al. 2015). In graph cut methods, data is treated as a header and
edging layout structure. The head is the top of each pixel, and adjacent pixels
are tied with a balanced edge based on their similarity. In cases involving
segmentation of multiple poster images, each poster also has a special sum-
mit called a station. Pixel heads are associated with all stations, where mark
weights determine the likelihood of classification. The way to find pieces in a
graph at the lowest cost is to get hash with the least amount of energy avail-
able, and the minimum flow algorithm will solve this problem. (Cho, Kim,
and Tai 2014; Fiss, Curless, and Szeliski 2015; Ferlay, Soerjomataram, Dikshit
et al. 2015; Chen, Lin, Yu, Kang, and Yu 2014; Ferlay, Soerjomataram, Dikshit
et al. 2015). The top of the growth is attached to one end after cutting, which
means placing the opposite poster on the other side. Our method also uses
the scheme-splitting process. These segmentation techniques are known as
moderate strategies since they are proof requiring client intercession. While
some realistic section strategies (Abdullah et al. 2012; Abdullah et al. 2013;
Saeed et al. 2019a; Saeed and Jafri 2015) can deal with information from any
separation, they are not always perfect for high-dimensional information, for
example, video clips. Video information has a conflicting structure along the
time hub, in contrast to 3D storage data. Therefore, fragmentation methods
can be strengthened by taking into account the dysfunctional neighborhood
relationships (Jarabo et al. 2014; D. Horn and Chen 2007). Video segmentation
quality can be improved by identifying live neighborhood relationships that
correspond to neighborhood frame pixels. A problem with clear 4D field data
that is similar to video data is that repetition is evident in the complex field
(Bishop, Tom E., and Paolo Favaro, 2007).
84 Machine Learning for Healthcare

Our research method is the first method that uses a graphic cut method
to focus on segmenting a 4D light field. Meanwhile, some unsupervised
approaches may be used for wide-angle or 4D focus images. Kolmogorov
and Zabih (2002) suggested dividing the 4D light fields based on a level-
ling method (Kolmogorov and Zabih 2002; Kowdle et al. 2012; Levoy and
Hanrahan 1996) that applies an active contour method to a large 4D por-
tion. The researchers Lin, Chen, Kang, and Yu (2015), suggest a method
for deep marking of multiple-width images based on the fact that fore-
ground objects cannot be excluded by deeper objects. Additionally, Saeed,
Abdullah, Jhanjhi, and Abdullah (2019b), Maeno et al. (2013), and Marx
et al. (2009) suggest a method to automatically extract objects from images
of different lengths using the contrast signal. The technique uses contrast
and appearance signals in multiple images to determine the probability of
foreground objects. Shaw et al. define areas of transparent species as bright
areas, and an uncensored approach is suggested. This method uses the
light field distortion function (Ng et al. 2005; Osher and Sethian 1988; Lin
et al. 2015, which represents the possibility that pixels belong to a trans-
parent object area, as well as the method for segmenting divided binary
graphics. Despite the success of these methods, the uncensored methods
are not suitable for clearly selecting the region for its liberation because
the areas of interest differ from one user to another, (Saeed and Abdullah
2019; Platt 1999; Rother, Kolmogorov, and Blake 2004). This suggests a seg-
mentation method for images using a 4D light field that uses presence and
contrast signals similar to (Wanner and Goldluecke 2012; Wanner, Meister,
and Goldluecke 2013) with respect to supervised methods. They train a
randomly chosen forest workbook to combine appearance and inequality
in order to deal with different types of information in nature and obtain
a specific probability for each brand. Despite its success, the only way
that segmentation results occur is in the two-dimensional central image
(Wanner, Straehle, and Goldluecke 2013; Xu, Nagahara, Shimada, and
Taniguchi 2015).

7.2.1 Static Structure of Literature Review with

Another Research Comparison
Table 7.1 shows the complete details of other research relating to the current
work of graph cutting as mention in previous section and introduction.

7.3 Methodology
The proposed research aims to collect data relating to the detection of brain
cancer (due to the creation of CSF leaks) with a high MRI interface. The
Comparison Analysis of Multidimensional Segmentation 85

TABLE 7.1
Comparison of Previous Research
Author’s
Name Affiliation Techniques Responses/Results
Brain Tumor Miller 2016 To Provide the information High Response time
(Cancer) of Cancer, Brain cancer,
spin and spinal cord
Brain tumor with Giorgio 2013 To Provide the details of High Response time,
MRI Deep Neural Network High Accuracy
Cerebrospinal Altaf et al. 2016 To Provide the detail High Response time
Fluid-CSF of CSF
Cerebrospinal Green et al. 2017 To Provide the High Response time
Fluid leakage experimental results of
CSF leakage
4D light field Mihara et al. To provide the Results of High Response time,
segmentation 2018 light field editing tools High Accuracy
method
Light Field Toolbox Mihara et al. To provide the Results of High Response time,
2018 light field editing tools High Accuracy
and Photoshop R

researcher discusses damaged brain cells caused by cell abnormalities. It is

a qualitative research study, therefore; it includes extensive primary and sec-
ondary sources, including comparative research studies from contemporary
publications. The main goal of our work is to build a framework that can rec-
ognize the field of tumors or isolate between malignant tumors and benign
tumors. Initially, the MRI image is outfitted with a specific final image fix
target for the rest of the procedures with the help of the high-dimensional
clear field toolbox and its graphical representation.
The sampling groups used for the study are medical brain images of
humans. The researcher will practically implement and simulate experi-
ments (3D and 4D image segmentation process) for tests on humans and
their random brain tissue (a sample from the human brain) from Malaysia.
The main goal of our proposed model is to detect the leakage point using
graph cutting that can detect cerebrospinal fluid loss in the brain and
tumor area, or that can differentiate between the patients who are with
and without tumors. The data is collected through various sources, which
apply the necessary research tools. The graph cutting tool is one of the
main tools which will be highlighted in the main point, and which will
be used to help identify the disease. In this research, graph cutting algo-
rithms explore the missing information by taking the attribute such as
stage of cancer and CSF leakage. This model identifies cancer at a very
early stage.
Figure 7.2 shows the structure of the methodological work as we follow the
direction of this method step by step and proposed the algorithm
86 Machine Learning for Healthcare

FIGURE 7.2
Conceptual framework of methodology.

FIGURE.7.3
(a) Binary transformation graph cutting tool in MATLAB.

7.3.1 Original Result of Image Testing in Binary Transformation

The binary transformation form of graph cutting is given below:
Figure 7.3 (a) shows the selected area of region in a skull and implements
the filter tool using the graph cutting tool in this image after the binary
transformation to refine the image.
Figure 7.3 (a) and (b) show the end stage of cancer (tumor) as this stage is
not recovered as it is clearly visible in the images. In this figure shows that
image is the last stage of cancer (tumor) with few forms of CSF leakage as
this is totally converted to the hard shell of the tumor as well as cancer which
is deposited in the upper side of the brain. After implementation of the graph
cutting method, it clearly displays the size of the tumor in 3D format. Grab-
cut is a graphic form of clip art. The algorithm estimates the black and white
distribution of the target object and the greyscale in the background using
the Gaussian model and starting with turning a user-defined chart in the
object into a clip. This is used to create a random Markov field in pixel tags
using the control function.
Comparison Analysis of Multidimensional Segmentation 87

7.3.2 High Dimension Structured Graphs

The explanation of high dimension graph structured is given below:

7.3.2.1 Grab-Cut
Grab-cut is a graphic form of clip art. The calculation assesses the shading
dissemination of the objective article and the foundation shading utilizing
the Gaussian mixture model and starts with turning a client characterized
pattern in the item into an area. This is utilized to make an arbitrary Markov
field in pixel labels by utilizing the control work. The incline is towards
a similar blurb in the associated zones and improves the graph to gather
their qualities. This gauge might be more exact than the first taken from the
limit outline. This methodology is rehashed in two stages until combination
happens. Clients of mistakenly grouped and improved areas can likewise
address gauges. The technique likewise amends the outcomes to protect the
edges.

7.4 Algorithm
The relevant algorithms are given below:

Algorithm.1: For Detection of CSF Spot in Image

(One Input and One Output-2D)

1. % Convert RGB DICOM into L*a*b* BW space.

2. X = rgb2lab (BW);
3. % Create vacant mask.
4. BW = false (size(X,1),size(X,2));
5. m = size(BW, 1);
6. n = size(BW, 2);
7. added Area = poly2mask(xPos, yPos, m, n);
8. BW = BW | added Area;
9. % Create concealedDICOM.
10.ConcealedDICOM (repmat(~BW,[1 1 3])) = 0;
11.End;

Algorithm.2: For Texture Feature Image Include

(One Input and One Output)

1.% Convert RGB DICOM into L*a*b* BW space.

2.X = rgb2lab(BW);
3.% Create vacantconcealment.
4.BW = false (size(X, 1), size(X, 2));
88 Machine Learning for Healthcare

5. m = size(BW, 1);
6. n = size(BW, 2);
7. Added Area = poly2mask (xPos, yPos, m, n);
8. % Create concealed DICOM.
9. Concealed DICOM = BW
10.Concealed DICOM (repmat(~BW,[1 1 3])) = 0;
11.End;

Algorithm.3: For Detection of CSF Spot in Image

(Two Outputs and One Input)

12.% Convert RGB DICOM into L*a*b* color space.

13.X = rgb2lab (RGB);
14.% Create vacantconcealment.
15.BW = false (size(X,1),size(X,2));
16.% Draw Freehand
17.m = size(BW, 1);
18.n = size(BW, 2);
19.Added Area = poly2mask (xPos, yPos, m, n);
20.BW = BW | added Area;
21.% Create concealed DICOM.
22.Concealed DICOM = RGB;
23.Concealed DICOM (repmat(~BW,[1 1 3])) = 0;
24.End;

7.5 Result Comparison and Discussion

These comparisons are based on semi-supervised machine learning as
we work on the comparison between 3D and segmentation of the dimen-
sional images. The findings of the tests are conducted by a dataset called
“Malignant Brain Cancer with CSF Leakage” by compiling previous and
current images using graph-cutting algorithms. Using the training data-
set, we trained several supervised machine-learning models. In this article,
the researcher reveals a brain cancer as we analyse both dimensions MRI
images and interfacing image field segmentation. Initially, the MRI han-
dled the prepared image method with the ultimate goal of adjusting the
image for the rest of the procedure. According to this study, brain cancer
and CSF were detected due to the interface of the 4D image segmentation
process.
Therefore, the research study consists of primary and secondary sources,
followed by a cutting technique for the software using graph-cutting tool that
uses images of original medical samples to measure the extent of damaged
Comparison Analysis of Multidimensional Segmentation 89

cells in the brain and applies the method of cutting graphics using MATLAB
tools and algorithms. In this study, the researcher proposes a 4D modula-
tion method that monitors the light field that can be used to emit light with
changing colour and binary. By creating a 4D hierarchy, 4D light fields can
be divided to reduce the graphics algorithm generated by the 4D scheme.
The researcher uses this technique to release damaged brain samples from
the brain’s skull. These results demonstrate the effectiveness of our approach
to light editing applications. Light field methods can be useful in improving
the quality of photo editing applications and compound lighting field tubes,
as they reduce the effects of artistic edges.
We aim to overcome the value of lost data in the computational experi-
ments of the proposed new method. To do this, we used a structured algo-
rithm of graphs with correlation and multiple types of data in time series.
We explained that improving the calculated data to determine the variables
of the verification method is related to the delay of the test time and the
training vector resulting from the time delay. The graph cut method demon-
strates the accuracy of images with results, and maintains color change and
binary change, as well as producing outputs and inputs.
Table.7.2 shows the comparison between the 3D and 4D (Figure 7.4.a,
Figure 7.4.b) dimension image segmentation processes, as we can see the
huge difference between the results and accuracy of the images. In the 4D
image, the researcher used a light field image segmentation process. In light
field segmentation process shows the two inputs and generates the output
in binary transformation after the implementation of the graph cutting tool
with a filter on the edge cutting image but in 4D only one input generates the
binary transformation in output form.

7.6 Conclusion
In this article, the researcher suggests that there is a supervised 4D field for
structured 4D graphics, and that 4D light field images can be divided using
a graphic clipping algorithm. Experimental results show that our method
achieves greater accuracy than previous methods using general lighting
field data sets from the American Cancer Society Center and CSF. In addi-
tion, the researcher applied the proposed method to the original images and
showed the result in changing the color and duo with one and two direc-
tors. These results demonstrate the efficiency of our clear field image editing
process using the graphics algorithm. The problem is calculation time dura-
tion. Because the use of the graphic cut algorithm requires a large amount
of calculation time when there are many peaks; the obvious future goal is to
solve this problem.
TABLE.7.2
90

Comparison of Previous and High Dimension Image

SNO. Images Previous Image High Dimension Image
1 Binary Transformation
image(part-1)

Figure 7.3a
Figure 7.4a
2 Binary Transformation
image(part-2)

Figure 7.4b
Machine Learning for Healthcare

Figure 7.3b
Comparison Analysis of Multidimensional Segmentation 91

Acknowledgments
The authors are grateful to the Department of Software Engineering, and the
faculty of computing, at Universiti Teknologi Malaysia-UTM, Malaysia for
the financial support needed to carry out this work.

References
Abdullah, Afnizanfaizal, et al. “An improved swarm optimization for parameter
estimation and biological model selection”. PLoS One 8(4) (2013): 4.
Abdullah, Afnizanfaizal, et al. “Cerebrospinal fluid pulsatile segmentation-a
review”. The 5th 2012 Biomedical Engineering International Conference, IEEE,
Thailand 2012.
Adelson, Edward H., and James R. Bergen. The Plenoptic Function and the Elements
of Early Vision, Vol. 2. Vision and Modeling Group, Media Laboratory,
Massachusetts Institute of Technology, Cambridge, MA: MIT Press, 1991.
Altaf I, Vohra AH and Shams S. “Management of cerebrospinal fluid leak following
posterior cranial fossa surgery”. Pak. J. Med. Sci. 2 (2016): 1439–1443.
Berent, Jesse, and Pier Luigi Dragotti “Unsupervised extraction of coherent regions
for image based rendering”. Proceedings of the British Machine Vision
Conference 2007, University of Warwick, BMVC, United kingdom, (2007).
Bishop, Tom E., and Paolo Favaro. “The light field camera: Extended depth of field,
aliasing, and superresolution”. IEEE Transactions on Pattern Analysis and Machine
Intelligence 34(5) (2011): 972–986.
Boykov, Yuri Y., and M.-P. Jolly. “Interactive graph cuts for optimal boundary &
region segmentation of objects in ND images”. Proceedings of the Eighth IEEE
International Conference on Computer Vision. ICCV 2001, Vol. 1. IEEE, Vancouver,
BC, Canada, 2001.
Boykov, Yuri, and Vladimir Kolmogorov. “An experimental comparison of min-cut/
max-flow algorithms for energy minimization in vision”. IEEE Transactions on
Pattern Analysis and Machine Intelligence 26(9) (2004): 1124–1137.
Boykov, Yuri, Olga Veksler, and Ramin Zabih. “Fast approximate energy minimiza-
tion via graph cuts”. IEEE Transactions on Pattern Analysis and Machine Intelligence
23(11) (2001): 1222–1239.
Chen, Can, et al. “Light field stereo matching using bilateral statistics of surface
cameras”. Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, San Juan, Puerto Rico, USA, 2014.
Cho, Donghyeon, Sunyeong Kim, and Yu-Wing Tai. “Consistent matting for light
field images”. European Conference on Computer Vision, Springer, Cham, 2014.
Ferlay, Jacques, et al. “Cancer incidence and mortality worldwide: Sources, methods
and major patterns in GLOBOCAN 2012.” International Journal of Cancer 136(5)
(2015): E359–E386.
Fiss, Juliet, Brian Curless, and Rick Szeliski. “Light field layer matting”. Proceedings of
the IEEE Conference on Computer Vision and Pattern Recognition, San Juan, Puerto
Rico, USA, 2015.
92 Machine Learning for Healthcare

Gamage, P. T., and Dr Lochandaka Ranathunga. “Identification of brain tumor using

image processing techniques”. Faculty of Information Technology, University of
Moratuwa, 2017. https://www. researchgate. net/publication/276133543.
Giorgio A, De Stefano N. “Clinical use of brain volumetry”. J Magn Reson Imaging
37(1) (2013): 1–55.
Green AL, et al. “A multicentre, prospective, randomized, controlled study to eval-
uate the use of a fibrin sealant as an adjunct to sutured dural repair”. Br. J.
Neurosurg. 2(2) (2014): 1–7.
Greig, Dorothy M., Bruce T. Porteous, and Allan H. Seheult. “Exact maximum a pos-
teriori estimation for binary images”. Journal of the Royal Statistical Society: Series
B (Methodological) 51(2) (1989): 271–279.
Hajime Mihara et al. “4D Light Field Segmentation with Spatial and Angular
Consistencies”. MEXT Japan, 2 (2018): 1–8.
Horn, Daniel Reiter, and Billy Chen. “Light shop: Interactive light field manipulation
and rendering”. Proceedings of the 2007 Symposium on Interactive 3D Graphics and
Games, New York, 2007.
Jarabo, Adrian, et al. “How do people edit light fields?” ACM Transactions on Graphics
(Proc. SIGGRAPH) 33(4): 1–12, (2014).
Kimberly Miller. “Cancer treatment & survivorship facts & figures”, American Cancer
Society 2(1) (2016): 1–44.
Kolmogorov, Vladimir, and Ramin Zabih. “Multi-camera scene reconstruction
via graph cuts”. European Conference on Computer Vision. Springer, Berlin,
Heidelberg, 2002.
Kowdle, Adarsh, Sudipta N. Sinha, and Richard Szeliski. “Multiple view object
cosegmentation using appearance and stereo cues.” European Conference on
Computer Vision. Springer, Berlin, Heidelberg, 2012.
Levoy, Marc, and Pat Hanrahan. “Light field rendering”. Proceedings of the 23rd Annual
Conference on Computer Graphics and Interactive Techniques, New York, 1996.
Lin, Haiting, et al. “Depth recovery from light field using focal stack symmetry”.
Proceedings of the IEEE International Conference on Computer Vision, France, 2015.
Marx, K., Andrew Lumsdaine, and Todor Georgiev. “The focused plenoptic camera”.
IEEE International Conference on Computational Photography (ICCP). IEEE, 2009.
Maeno, Kazuki, et al. “Light field distortion feature for transparent object recog-
nition”. Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, San Juan, Puerto Rico, 2013.
Mendrik, Adriënne M., et al. “MRBrainS challenge: Online evaluation framework
for brain image segmentation in 3T MRI scans”. Computational Intelligence and
Neuroscience 2015, 1–12, (2015).
Nagahashi, Tomoyuki, Hironobu Fujiyoshi, and Takeo Kanade. “Video segmentation
using iterated graph cuts based on spatio-temporal volumes”. Asian Conference
on Computer Vision. Springer, Berlin, Heidelberg, 2009.
Ng, Ren, et al. “Light field photography with a hand-held plenoptic camera”. Computer
Science Technical Report CSTR 2(11) (2005): 1–11.
Osher, Stanley, and James A. Sethian. “Fronts propagating with curvature-depen-
dent speed: Algorithms based on Hamilton-Jacobi formulations”. Journal of
Computational Physics 79(1) (1988): 12–49.
Platt, John. “Probabilistic outputs for support vector machines and comparisons to
regularized likelihood methods”. Advances in Large MARGIN Classifiers 10(3)
(1999): 61–74.
Comparison Analysis of Multidimensional Segmentation 93

Rother, Carsten, Vladimir Kolmogorov, and Andrew Blake. “’GrabCut’ interactive

foreground extraction using iterated graph cuts.” ACM Transactions on Graphics
(TOG) 23(3) (2004): 309–314.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi. “Analysis of the lung can-
cer patient’s for data mining tool”. IJCSNS 19(7) (2019a): 90.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi. “Implementation of
Fourier transformation with brain cancer and CSF images”. Indian Journal of
Science and Technology 12(37) (2019b): 37.
Saeed, Soobia, and Afnizanfaizal Abdullah. “Recognition of brain cancer and cere-
brospinal fluid due to the usage of different MRI image by utilizing support
vector machine”. Bulletin of Electrical Engineering and Informatics 9(2) (2020):
619–625.
Saeed, Soobia, and Afnizanfaizal Bin Abdullah. “Investigation of a brain cancer with
interfacing of 3-dimensional image processing”. 2019 International Conference on
Information Science and Communication Technology (ICISCT). IEEE, Pakistan, 2019.
Saeed, Soobia, and Raza Jafri. “Estimation of brain tumor using latest technology of
mobile phone”. Journal of Information and Communication Technology (JICT) 9(1)
(2015): 32–09.
Wanner, Sven, and Bastian Goldluecke. “Globally consistent depth labeling of 4D
light fields”. 2012 IEEE Conference on Computer Vision and Pattern Recognition.
IEEE, Rhode Island, 2012.
Wanner, Sven, Stephan Meister, and Bastian Goldluecke “Datasets and benchmarks
for densely sampled 4D light fields”1–9. VMV 13 (2013).
Wanner, Sven, Christoph Straehle, and Bastian Goldluecke. “Globally consistent
multi-label assignment on the ray space of 4d light fields”. Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition, Washington, D.C.,
2013.
Xu, Yichao, et al. “Transcut: Transparent object segmentation from a light-field
image”. Proceedings of the IEEE International Conference on Computer Vision,
France, 2015.
8
Deep Convolutional Network Based
Approach for Detection of Liver Cancer
and Predictive Analytics on Cloud

Pramod H. B. and Goutham M.

CONTENTS
8.1 Introduction................................................................................................... 95
8.1.1 Types of Liver Diseases.................................................................... 98
8.2 Medical Images and Deep Learning........................................................ 100
8.2.1 Micro-Service Architecture........................................................... 101
8.2.2 Integration of NVDIA GPU for Deep Learning on Cloud........ 101
8.2.3 Presenting the Sockets and Slots for Processors........................ 102
8.2.4 Clock Details of Deep Learning Server....................................... 102
8.2.5 Threads for Deep Learning–Based Computations.................... 103
8.2.6 Available Hard Disk for Use......................................................... 103
8.2.7 Memory............................................................................................ 103
8.2.8 Overall Details of Used Computing Environment with
Deep Convolutional Networks..................................................... 103
8.3 Deep Learning for Liver Diagnosis with the Projected Model............ 103
8.4 Proposed Model and Outcomes............................................................... 104
8.5 Conclusion................................................................................................... 107
References.............................................................................................................. 109

8.1 Introduction
Health is one of the most important things for human beings and thus it needs
to be maintained with the utmost care. Since the advent of human life, the
wish to live better and for longer has been of great interest to researchers. A
number of techniques and approaches in the medical sciences exist whereby
the enormous paradigms and theories relate to the understanding of human
life (Ferlay et al. 2010; Lu et al. 2006; Moghbel et al. 2018; Prasoon et al. 2013).
This chapter focuses on the advanced implementation of deep learning
based on convolutional networks for medical diagnosis, and it places a spe-
cific focus on diseases of the liver as the liver is one of the key components
95
96 Machine Learning for Healthcare

and organs within the human body (Ronneberger et al. 2015; Stollenga et al.
2015; Roth et al. 2015; Wang et al. 2015).
Nowadays, the world faces an enormous number of diseases of which
many are life-threatening. Some organs are considered as particularly sensi-
tive as far as dangerous diseases are concerned, and these are:

1. Liver
2. Heart
3. Kidney
4. Lungs

If even a small segment of a hazardous disease touches these organs, it will

be quite dangerous for the person’s life.
Life-threatening diseases are chronic, often incurable diseases, which have
the effect of considerably limiting a person’s life expectancy (Li et al. 2016;
Çiçek et al. 2016; Havaei et al. 2017; Chen et al. 2017). These include cancer,
diabetes, neurological conditions, coronary heart disease, and HIV/Aids.
Others include:

1. Dengue
2. Ebola
3. Plague
4. Enterovirus
5. Cholera
6. MRSA
7. Cerebrovascular
8. Chagas
9. Meningococcal
10. Necrotizing

Liver disease occurs as a result of infections in different ways as mentioned here:

1. Hepatitis A: Fatigue, Diarrhea, Fever, Nausea

2. Hepatitis B: Weakness, Pain, Yellowing
3. Hepatitis C: Muscle Aches, Yellow Tinge, Bowel Movements

The common symptoms of the liver disease are:

1. Dark urine
2. Yellow skin and eyes, known as jaundice
3. Vomiting
4. Pale, bloody, or black stool
Predictive Analytics on Cloud 97

5. Nausea
6. Swollen ankles, legs, or abdomen
7. Itchy skin
8. Easy bruising
9. Decreased appetite
10. Ongoing fatigue
11. Jaundice
12. Diarrhea
13. Confusion
14. Fatigue and weakness
15. Nausea

The following are the key risk factors associated with liver issues:

1. Getting a tattoo or body piercing with non-sterile needles

2. Sharing needles
3. Having a job where you’re exposed to blood and other bodily fluids
4. Unsafe sex
5. Diabetes
6. High cholesterol
7. Having a family history of liver disease
8. Being overweight
9. Exposure to toxins or pesticides
10. Taking certain supplements or herbs, especially in large amounts
11. Ascites
12. Accumulation of fluid in the abdomen
13. Endoscopy
14. Nonsurgical procedure
15. Jaundice
16. Yellowing of the skin and the whites of the eyes
17. Tumors
18. Abnormal growth of cells
19. Varices
20. Shunts

Various issues that affect the human liver can be identified in a CT Scan or
through deep images of the liver, with medical imaging and dynamic librar-
ies used for getting the images.
98 Machine Learning for Healthcare

These issues include:

1. Lesions
2. Injuries
3. Bleeding
4. Infections
5. Abscesses
6. Infections
7. Obstructions

In terms of the cumulative health of the human liver, fatty liver cells with
less than 5% are considered to be healthy. If these values move beyond 5%, it
is considered as a fatty liver which is one of the key diseases throughout the
world nowadays.
Statista is one of the key portals for research and statistical analytics
whereby the enormous research surveys and development are discussed
with a high degree of accuracy and presentation. In the following research
analytics, liver transplants are presented. There have been huge related ele-
vations conducted in the United States in 2018, and it must be taken into
consideration by the researchers as liver disease is being research using deep
learning (Wolpert 1992; Soler et al. 2001; Moltz et al. 2008; Wong et al. 2008).
The statistical analytics present the major locations of key countries that
are suffering from health issues and liver problems, as shown in graphical
outcomes in terms of ethnicity.
Figure 8.1 highlights the assorted stages of liver issues which need to be
evaluated and understood while presenting a predictive analytics on liver
disease.

8.1.1 Types of Liver Diseases

1. Laennec’s cirrhosis
2. Hepatic encephalopathy
3. Non-cirrhotic portal fibrosis
4. Hy’s law
5. Peliosis hepatis
6. Liver failure
7. Wilson’s disease
8. Liver abscesses
9. Epithelial-mesenchymal transition
10. Bland embolization
11. Alcoholic liver disease
Predictive Analytics on Cloud 99

FIGURE 8.1
Stages of liver issues.

12. Cirrhosis
13. Hepatosplenomegaly
14. Chronic liver disease
15. Zahn infarct
16. Fibrolamellar hepatocellular carcinoma
17. Alpha-1 antitrypsin deficiency
18. North American Indian childhood cirrhosis
19. Hepatopulmonary syndrome
20. Gastric antral vascular ectasia
21. Focal fatty liver
22. Congenital hepatic fibrosis
23. Acute liver failure
24. Congenital hypofibrinogenemia
25. Fatty liver disease
26. Primary biliary cholangitis
27. Pediatric end-stage liver disease
28. Hepatolithiasis
29. Hepato-biliary diseases
30. Viral hepatitis
31. Hepatotoxicity
100 Machine Learning for Healthcare

32. Mucinous cystic neoplasm

33. Acute fatty liver of pregnancy
34. Alveolar hydatid disease
35. Congestive hepatopathy
36. Steatohepatitis
37. Polycystic liver disease
38. Indian childhood cirrhosis
39. Bacillary peliosis
40. Liver cancer
41. Progressive familial intrahepatic cholestasis
42. Non-alcoholic fatty liver disease
43. Zieve’s syndrome

All of these illnesses and diseases need to be analyzed with effective

algorithms.

8.2 Medical Images and Deep Learning

There is a very close relationship between Artificial Intelligence (AI),
machine learning, deep learning, convolutional networks, and deep learn-
ing–based convolutional networks. These paradigms are widely used for
solving the high-performance computations associated with medical imag-
ing and many other related challenges (Jimenez-Carretero et al. 2011; Huang
et al. 2014; Vorontsov et al. 2014; Le et al. 2016; Kuo et al. 2017). Figure 8.2
presents the close association between the terms AI, machine learning, and
deep learning which are widely integrated for data analytics in the medical
domain.
All these research streams are closely associated and when used together
from the highest degree of performance and computational intelligence.
The usage patterns of these approaches can be extracted from the enormous
implementation patterns which are used in assorted research domains
and even in cloud-based platforms and research environments (33Conze
et al. 2017; Hoogi et al. 2017; Chaieb et al. 2017; Christ et al. 2017; Han 2017;
Vorontsov et al. 2017).
The following are the tools and technologies used for the implementation
of deep learning–based algorithms in cloud-based environments so that
there is no need to purchase the dedicated infrastructure (Table 8.1).
The key features embedded in the suites include: (Chlebus et al. 2017; Zhou
et al. 2017a; Farag et al. 2017; Zhou et al. 2017; Roth et al. 2017).
Predictive Analytics on Cloud 101

FIGURE 8.2
Association in assorted segments of Artificial Intelligence.

8.2.1 Micro-Service Architecture

1. APIs for Java, Python, and Scala

2. Massive amounts of data processed using clusters
3. Scalability on Hadoop for big data
4. Parallel computing and training
5. Distributed architecture with multi-threading
6. Support for GPU for scalability on Amazon Web Services (AWS)
cloud
7. Support to CPU and GPU (Table 8.2)

8.2.2 Integration of NVDIA GPU for Deep Learning on Cloud

!nvidia-smi
!nvidia-smi -L
!lscpu | grep “Model”
102 Machine Learning for Healthcare

TABLE 8.1
Prominent Free and Open Source Tools for Deep Learning
Tool / Library URL
Apache Mahout https://mahout.apache.org/
DLib http://dlib.net/
Apache Singa https​://si​nga.i​ncuba​tor.a​pache​.org/​en/in​dex.h​tml
Shogun http://www.shogun-toolbox.org/
Scikit-Learn http://scikit-learn.org/
OpenNN http://www.opennn.net/
Chainer https://chainer.org/
OpenAI https://openai.com/
Edward http://edwardlib.org/
Microsoft Cognitive Toolkit https​://ww​w.mic​rosof​t.com​/en-u​s/cog​nitiv​e-too​lkit/​
DeepLearning4j https://deeplearning4j.org/
Apache Spark MLib https://spark.apache.org/mllib/
Lime https://github.com/marcotcr/lime
PyTorch http://pytorch.org/
Torch http://torch.ch/
MXNet https://mxnet.apache.org/
Neon http://neon.nervanasys.com
TensorFlow https://www.tensorflow.org/
Caffe http://caffe.berkeleyvision.org/
Gensim https://radimrehurek.com/gensim/
Oryx 2 http://oryx.io/
MLDB https://mldb.ai/
Keras https://keras.io/

TABLE 8.2
Cloud-Based Deep Learning Services for High Performance Computations
Cloud Service URL
Google Colaboratory http://colab.research.google.com
PaperSpace https://www.paperspace.com
Neptune https://www.neptune.ml
Nvidia GPU Cloud https://www.nvidia.com/en-us/gpu-cloud
BigML https://www.bigml.com
GPU Eater https://www.gpueater.com

8.2.3 Presenting the Sockets and Slots for Processors

!lscpu | grep “Details of Socket”
!lscpu | grep “Number of Cores / Socket:”

8.2.4 Clock Details of Deep Learning Server

!lscpu | grep “MHz”
Predictive Analytics on Cloud 103

8.2.5 Threads for Deep Learning–Based Computations

!lscpu | grep “L3 Cache”

!lscpu | grep “Thread(s) per Core”

8.2.6 Available Hard Disk for Use

!df –hT /

8.2.7 Memory
!cat /proc/meminfo | grep “MemAvailable”

8.2.8 Overall Details of Used Computing Environment

with Deep Convolutional Networks

1. CPU
• Xeon Processors
• 45MB Cache
• 1 Single core Hyper Threaded (1 Core, 2 Threads)
• 3Ghz (No Turbo Boost)
2. GPU
• 2496 CUDA cores
• Compute 3.7
• 1xTesla K80
• 12GB GDDR5 VRAM
• Server Idle Time: 90 minutes
• RAM: More than 12 GB on Cloud
• Disk: More than 30 GB on Cloud

8.3 Deep Learning for Liver Diagnosis

with the Projected Model
Images of the liver are taken by medical imaging and automation libraries.
The predicted mask has the association and knowledge of the key points and
features which are required for predictive analytics. The images are captured
by medical images and other IT-based tools. The neural network is formed
and thereby the dynamic network is formed as can be seen in Figure 8.3.
104 Machine Learning for Healthcare

FIGURE 8.3
Prediction of mask from the vessel image.

The key segments and phases in the projected approach include the
following:

1. Initial convolutional layers

2. Inception-resnet-A block
3. Reduction
4. Inception-reset-B block
5. Reduction
6. Inception-resnet-C block
7. Average Pooling
8. Towards the neural features

The models mentioned herein refer to the pre-trained models which are
prominently used for the deep learning–based environment. These models
are quite effectual in achieving higher performance accuracy as they have
the key features in the repository with pre-trained models and the learning
environment for the predictive analytics as presented in Figure 8.4.
The training data has the image dataset with the contouring data. In
addition, the reinforcement learning is used so that the error factors can be
reduced in integration of the convolutional and recurrent neural networks as
depicted in Figure 8.5. The automatic contouring results are obtained with
the focus on the higher degree of performance.
The training datasets and the testing datasets are the key components in
the deep learning based–environment, as these are required for the effectual
predictive analytics as presented in Figure 8.6.

8.4 Proposed Model and Outcomes

The proposed model includes the training of the datasets of the Cancer
genome that makes up the benchmark dataset and is under research by
Predictive Analytics on Cloud 105

FIGURE 8.4
Flow of the network formation and further analytics.

enormous numbers of research groups. The assorted research groups are

making use of the dataset for the training and testing and validation of the
datasets, including for the approaches of machine learning, deep learning
and many others high performance integrations and presented in Figure 8.7.
epoch train_loss valid_loss accuracy
1 0.834818 1.048996 0.616667 (20:21)
CPU times: user 3min 21s, sys: 46.6 s, total: 4min 8s
Wall time: 20min 34s
Total time: 32:52
106 Machine Learning for Healthcare

FIGURE 8.5
Key points and phases.

FIGURE 8.6
Training and testing datasets with deep learning aspects.

epoch train_loss valid_loss accuracy

1 0.189783 0.225980 0.918750 (10:32)

2 0.162783 0.235727 0.912500 (11:26)
3 0.139040 0.232577 0.918750 (10:53)

Figure 8.8 depicts the flow of projected work for the above research approach
and thereby can be used to gather analytical outcomes (Table 8.3).
Predictive Analytics on Cloud 107

FIGURE 8.7
Training and testing datasets with the related aspects of deep learning.

The results obtained as in Figure 8.9, from the deep learning–based con-
volutional neural network–based results, presents that the effectiveness in
the outcomes can be achieved with a higher degree of performance and
effectiveness.

8.5 Conclusion
Medical diagnosis is very important process in which there is a need to inte-
grate high performance algorithms. Liver or Hepatic Cancer is one of the
riskiest disorders in which the continuance problem arises. In this chapter,
the dataset of liver cancer from the appraisal bit of Cancer Imaging Archive
is set up in the model of significant getting the hang of having multilay-
ered neural frameworks so the model can be set up with the dataset. The
presented scenario has the tweaking and trade work with the source work
and the affirmed layers of the neural framework so the monstrous evalua-
tion of the key concentrations and features of dangerous development can
be removed. The work is having the parameters for the appraisal of results
in different centers including precision, execution time, cost factors, and the
multifaceted nature with the objective that the total execution of the pro-
posed system should be conceivable. The work has joint significant learn-
ing with the convolutional frameworks concerning the grounds that there
are minute and exact zone extractions from the photos of liver infection for
dealing with and the vivacious mining. With the tremendous evaluation of
the features and related key impressions from the liver peril pictures, the
test data or beginning late passed on liver pictures can be empowered and
further accuracy segments can be found on the probabilities of thought of
108 Machine Learning for Healthcare

FIGURE 8.8
Flow of proposed work.

infection. The proposed approach is depended upon with the promise to the
prosperity sciences and regardless of for the space of telemedicine in which
the remote ID and treatment of the challenging issue is required. The sig-
nificant neural framework–based system is executed using the open source
suites of TensorFlow and Keras, so the status and testing should be conceiv-
able with a more raised degree of accuracy on the benchmark datasets of
Predictive Analytics on Cloud 109

TABLE 8.3
Analytics of Accuracy Levels
Scenario Accuracy: Traditional Aspects Accuracy: Projected Approach
1 71 97
2 70 94
3 71 93
4 73 93
5 71 93
6 79 91

FIGURE 8.9
Analytics of the outcomes.

liver cancer. Notwithstanding the convolutional neural framework, the addi-

tional layers with the tweaking factors are used so the general precision of
the check is raised with reliably raised degree of need and least destruction
up or botch rate and more significant levels of accuracy. The chapter presents
the accuracy levels associated with the cumulative performance for liver dis-
ease diagnosis using deep convolutional networks.

References
Rashmi Agrawal, “Predictive analysis of breast cancer using machine learning tech-
niques”. Ingeniería Solidaria, 15(29), pp. 1–23, 2019.
I. Ben-Cohen, E. Diamant, M. Klang, Amitai, and H. Greenspan, “Fully convolu-
tional network for liver segmentation and lesions detection”. In: International
Workshop on Large-Scale Annotation of Biomedical Data and Expert Label Synthesis,
Springer, pp. 77–85, 2016. Athens, Greece.
110 Machine Learning for Healthcare

J. Cai, L. Lu, Y. Xie, F. Xing, and L. Yang, “Improving deep pancreas segmentation
in ct and mri images via recurrent neural contextual learning and direct loss
function”. arXiv Preprint ArXiv:1707.04912, 2017.
F. Chaieb, T. B. Said, S. Mabrouk, and F. Ghorbel, “Accelerated liver tumor segmenta-
tion in four-phase computed tomography images”. Journal of Real-Time Image
Processing, 13(1), pp. 121–133, 2017.
H. Chen, Q. Dou, L. Yu, J. Qin, and P.-A. Heng, “Voxresnet: Deep voxelwise residual
networks for brain segmentation from 3D MR images”. NeuroImage, 2017. 170,
446–455.
H. Chen, D. Ni, J. Qin, S. Li, X. Yang, T. Wang, and P. A. Heng, “Standard plane local-
ization in fetal ultrasound via domain transferred deep neural networks”. IEEE
Journal of Biomedical and Health Informatics, 19(5), pp. 1627–1636, 2015.
G. Chlebus, H. Meine, J. H. Moltz, and A. Schenk, “Neural network-based automatic
liver tumor segmentation with random forest-based candidate filtering.” arXiv
Preprint ArXiv:1706.00842, 2017.
P. F. Christ, et al., “Automatic liver and lesion segmentation in ct using cascaded
fully convolutional neural networks and 3d conditional random fields”. In:
International Conference on Medical Image Computing and Computer-Assisted
Intervention, Springer, pp. 415–423, 2016. Athens, Greece.
P. F. Christ, et al., “Automatic liver and tumor segmentation of ct and mri vol-
umes using cascaded fully convolutional neural networks”. arXiv Preprint
ArXiv:1702.05970, 2017.
Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ronneberger, “3D u-net:
Learning dense volumetric segmentation from sparse annotation”. In:
International Conference on Medical Image Computing and Computer- Assisted
Intervention, Springer, 424–432, 2016. Athens, Greece.
P.-H. Conze, V. Noblet, F. Rousseau, F. Heitz, V. de Blasi, R. Memeo, and P. Pessaux,
“Scale-adaptive supervoxelbased random forests for liver tumor segmenta-
tion in dynamic contrast-enhanced ct scans”. International Journal of Computer
Assisted Radiology and Surgery, 12(2), 223–233, 2017.
Q. Dou, H. Chen, Y. Jin, L. Yu, J. Qin, and P.-A. Heng, “3d deeply supervised network
for automatic liver segmentation from ct volumes”. In: International Conference
on Medical Image Computing and Computer-Assisted Intervention, Springer, 149–
157, 2016. Athens, Greece.
L. Farag, H. R. Roth Lu, J. Liu, E. Turkbey, R. M. Summers, R. M. Summers, “A bottom-
up approach for pancreas segmentation using cascaded superpixels and (deep)
image patch labeling”. IEEE Transactions on Image Processing: A Publication of the
IEEE Signal Processing Society, 26(1), 386–399, 2017.
Jacques Ferlay, H. R. Shin, F. Bray, D. Forman, C. Mathers, D. M. Parkin, “Estimates of
worldwide burden of cancer in 2008: GLOBOCAN 2008”. International Journal of
Cancer, 127(12), 2893–2917, 2010.
X. Han, “Automatic liver lesion segmentation using a deep convolutional neural net-
work method”. arXiv Preprint ArXiv:1704.07239, 2017.
M. Havaei, A. Davy, D. Warde-Farley, A. Biard, A. Courville, Y. Bengio, C. Pal, P.-M.
Jodoin, and H. Larochelle, “Brain tumor segmentation with deep neural net-
works”. Medical Image Analysis, 35, 18–31, 2017.
C. F. Hoogi, G. M. Beaulieu, E. Heba Cunha, C. B. Sirlin, S. Napel, D. L. Rubin, D. L.
Rubin, “Adaptive local window for level set segmentation of CT and MRI liver
lesions”. Medical Image Analysis, 37, 46–55, 2017.
Predictive Analytics on Cloud 111

G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, “Densely connected con-
volutional networks”. In: Proceedings of the IEEE Conference on Computer Vision
and Pattern Recognition, 2017. Athens, Greece.
W. Huang, Y. Yang, Z. Lin, G.-B. Huang, J. Zhou, Y. Duan, and W. Xiong, “Random
feature subspace ensemble based extreme learning machine for liver tumor
detection and segmentation”. In: Engineering in Medicine and Biology Society
(EMBC), 2014 36th Annual International Conference of the IEEE, IEEE, 4675–4678,
2014. Chicago, Illinois.
D. Jimenez-Carretero, L. Fernandez-de Manuel, J. Pascau, J. M. Tellado, E. Ramon,
M. Desco, A. Santos, and M. J. Ledesma-Carbayo, “Optimal multiresolu-
tion 3d level-set method for liver segmentation incorporating local curva-
ture constraints”. In: Engineering in Medicine and Biology Society, EMBC, 2011
Annual International Conference of the IEEE, IEEE, pp. 3419–3422, 2011 Boston,
Massachusetts.
C.-L. Kuo, S.-C. Cheng, C.-L. Lin, K.-F. Hsiao, and S.- H. Lee, “Texture-based treat-
ment prediction by automatic liver tumor segmentation on computed tomog-
raphy”. In: Computer, Information and Telecommunication Systems (CITS), 2017
International Conference on. IEEE, pp. 128–132, 2017. Paris, France.
T.-N. Le, et al., “Liver tumor segmentation from mr images using 3D fast march-
ing algorithm and single hidden layer feedforward neural network”. BioMed
Research International, 2016, 1–8.
X. Li, Q. Dou, H. Chen, C.-W. Fu, and P.-A. Heng, “Multi-scale and modality dropout
learning for intervertebral disc localization and segmentation”. In: International
Workshop on Computational Methods and Clinical Applications for Spine Imaging,
Springer, pp. 85–91, 2016. Athens, Greece.
X. Li, Q. Dou, H. Chen, C.-W. Fu, X. Qi, D. L. Belav`y, G. Armbrecht, D. Felsenberg,
G. Zheng, and P.- A. Heng, “3d multi-scale fcn with random modality voxel
dropout learning for intervertebral disc localization and segmentation from
multi-modality mr images”. Medical Image Analysis, 2018, 3419–3422.
Rui Lu, Pina Marziliano, and Choon Hua Thng, “Liver tumor volume estimation
by semi-automatic segmentation method”. In: IEEE Engineering in Medicine and
Biology 27th Annual Conference, IEEE, p. 2006, 2005.
F. Lu, F. Wu, P. Hu, Z. Peng, and D. Kong, “Automatic 3d liver location and segmenta-
tion via convolutional neural network and graph cut”. International Journal of
Computer Assisted Radiology and Surgery, 12(2), pp. 171–182, 2017.
Mehrdad Moghbel, S. Mashohor, R. Mahmud, M. I. B. Saripan, “Review of liver seg-
mentation and computer assisted detection/diagnosis methods in computed
tomography”. Artificial Intelligence Review, 50(4), pp. 497–537, 2018.
J. H. Moltz, L. Bornemann, V. Dicken, and H. Peitgen, “Segmentation of liver metas-
tases in ct scans by adaptive thresholding and morphological processing”. In:
MICCAI Workshop, 41(43) 195, 2008.
Adhish Prasoon, et al., “Deep feature learning for knee cartilage segmentation
using a triplanar convolutional neural network”. In: International Conference on
Medical Image Computing and Computer-Assisted Intervention, Springer, Berlin,
Heidelberg, 2013.
Olaf Ronneberger, Philipp Fischer, and Thomas Brox, “U-net: Convolutional net-
works for biomedical image segmentation”. In: International Conference on
Medical image Computing and Computer-Assisted Intervention. Springer, Cham,
2015.
112 Machine Learning for Healthcare

Holger R. Roth, et al., “Deeporgan: Multi-level deep convolutional networks for auto-
mated pancreas segmentation”. In: International Conference on Medical Image
Computing and Computer-Assisted Intervention, Springer, Cham, 2015.
H. R. Roth, L. Lu, N. Lay, A. P. Harrison, A. Farag, A. Sohn, and R. M. Summers, “Spatial
aggregation of holistically-nested convolutional neural networks for automated
pancreas localization and segmentation”. arXiv Preprint ArXiv:1702.00045, 2017.
K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale
image recognition”. arXiv Preprint ArXiv:1409.1556, 2014.
L. Soler, H. Delingette, G. Malandain, J. Montagnat, N. Ayache, C. Koehl, O. Dourthe,
B. Malassagne, M. Smith, D. Mutter, J. Marescaux, “Fully automatic anatomical,
pathological, and functional segmentation from ct scans for hepatic surgery”.
Computer Aided Surgery: Official Journal of the International Society for Computer
Aided Surgery, 6(3), 131–142, 2001.
Marijn F. Stollenga, et al., “Parallel multi-dimensional lstm, with application to fast
biomedical volumetric image segmentation”. Advances in Neural Information
Processing Systems, 2015. 2998–3006.
C. Sun, S. Guo, H. Zhang, J. Li, M. Chen, S. Ma, L. Jin, X. Liu, X. Li, and X. Qian,
“Automatic segmentation of liver tumors from multiphase contrast-enhanced
CT images based on fcns”. Artificial Intelligence in Medicine, 2017. 83, 58–66.
N. Tajbakhsh, J. Y. Shin, S. R. Gurudu, R. T. Hurst, C. B. Kendall, M. B. Gotway, and J.
Liang, “Convolutional neural networks for medical image analysis: Full train-
ing or fine tuning?” IEEE Transactions on Medical Imaging, 35(5), 1299–1312, 2016.
K.-L. Tseng, Y.-L. Lin, W. Hsu, and C.-Y. Huang, “Joint sequence learning and
cross-modality convolution for 3d biomedical segmentation”. arXiv Preprint
ArXiv:1704.07754, 2017.
Z. Tu, “Auto-context and its application to high-level vision tasks”. In: Computer Vision
and Pattern Recognition, 2008. CVPR 2008. IEEE Conference on. IEEE, 1–8, 2008.
E. Vorontsov, N. Abi-Jaoudeh, and S. Kadoury, “Metastatic liver tumor segmenta-
tion using texture-based omni-directional deformable surface models”. In:
International MICCAI Workshop on Computational and Clinical Challenges in
Abdominal Imaging, Springer, 74–83, 2014. Paris, France.
E. Vorontsov, G. Chartrand, A. Tang, C. Pal, and S. Kadoury, “Liver lesion segmenta-
tion informed by joint liver segmentation”. arXiv Preprint ArXiv:1707.07734, 2017.
J. Wang, J. D. MacKenzie, R. Ramachandran, and D. Z. Chen, “Detection of glands
and villi by collaboration of domain knowledge and deep learning”. In:
International Conference on Medical Image Computing and Computer- Assisted
Intervention, Springer, 20–27, 2015.
X. Wang, Y. Zheng, L. Gan, X. Wang, X. Sang, X. Kong, and J. Zhao, “Liver segmen-
tation from CT images using a sparse priori statistical shape model (sp-ssm)”.
PLoS One, 12(10), p. e0185249, 2017.
D. H. Wolpert, “Stacked generalization”. Neural netIEEE Transactions on Medical
Imaging 12 Works, 5(2), 241–259, 1992.
D. Wong, J. Liu, Y. Fengshou, Q. Tian, W. Xiong, J. Zhou, Y. Qi, T. Han, S. Venkatesh,
and S.-c. Wang, “A semi-automated method for liver tumor segmentation
based on 2d region growing with knowledge-based constraints”. In: MICCAI
Workshop, 41(43), 159, 2008.
Predictive Analytics on Cloud 113

Y. Zhou, L. Xie, E. K. Fishman, and A. L. Yuille, “Deep supervision for pancreatic cyst
segmentation in abdominal CT scans”. In: International Conference on Medical
Image Computing and Computer-Assisted Intervention, Springer, 222–230, 2017a.
Y. Zhou, L. Xie, W. Shen, Y. Wang, E. K. Fishman, and A. L. Yuille, “A fixed-point
model for pancreas segmentation in abdominal CT scans”. In: International
Conference on Medical Image Computing and Computer- Assisted Intervention,
Springer, 693–701, 2017b. Quebec City, Canada.
9
Performance Analysis of Machine
Learning Algorithm for Healthcare Tools
with High Dimension Segmentation

Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi,

Memood Naqvi and Mamoona Humayun

CONTENTS
9.1 Introduction................................................................................................. 115
9.2 Literature Review....................................................................................... 117
9.3 Methodology............................................................................................... 118
9.3.1 Proposed Framework..................................................................... 118
9.3.2 Light Field Toolbox for MATLAB................................................. 119
9.3.3 High Dimensional Light Field Segmentation Method.............. 119
9.3.4 High Dimensional Structured Graphs........................................ 119
9.4 High Dimension Structured Graphs........................................................ 119
9.4.1 Grab-Cut........................................................................................... 119
9.4.2 Image Testing Value....................................................................... 120
9.4.3 Image Testing Result...................................................................... 120
9.4.4 Graph Cut Value for B/W Image ................................................. 121
9.4.5 Image Testing Value....................................................................... 121
9.4.6 Image Testing Result...................................................................... 121
9.5 Algorithm..................................................................................................... 123
9.6 Result and Discussion................................................................................ 124
9.7 Conclusion................................................................................................... 125
9.8 Future Work................................................................................................. 125
Acknowledgment................................................................................................. 125
References.............................................................................................................. 126

9.1 Introduction
Graph cutting option in Image Segmented Image cutting program, Graph
Cut is a semi-automatic segmentation technology that can be used by a
researcher to separate an image into front and back components. We can
draw lines in the image, called scribbles, to determine what you want in the
115
116 Machine Learning for Healthcare

foreground and what you want in the background. The split image automati-
cally divides the image based on its scribble and displays the split image.
The researcher can improve fragmentation by drawing more scrawls on the
image to be satisfied with the result.
Graph Cut technology applies graphics theory to image processing to
achieve rapid fragmentation. The technique creates a graphic for the image
where each pixel is a knot connected to a weighted edge. The more likely the
pixel is bound the more weight. The algorithm cuts the weak edges, splitting
the objects in the image. The split image uses a specific set of the Graph Cut
algorithm called Slow Adjust. The information about graphics segmentation
on segmentation technology, such as local graphics segmentation “grab-cut”
is an image clip.
The integrated graphics clipping algorithms have been successfully applied
to a wide range of vision and graphics problems. This article focuses on the
simplest graphic cutting app: segmenting an object in image data. In addi-
tion to its simplicity, this application embodies the best features of harmonic
graphics cutting methods of vision: the optimum global level, practical effi-
ciency, numerical durability, the ability to integrate a wide range of signals,
optical restrictions, untied topological character sectors and the applicability
of ND problems (Boykov et al. 2015, Adelson and Bergen 1991, Berent and
Dragotti 2007, Bishop and Favaro 2011, Boykov and Jolly 2001).
Figure 9.1 shows that introducing a general concept for the first time to use
binary graphic cutting algorithms to split and verify objects using MATLAB
programming software, this idea was extensively studied in drawing and
computer vision societies. We provide links to an extensive variety of recog-
nized additions based on iterative reassessment, parameter getting to know,
multi-domain procedures, hierarchy, slender tiers, and different technolo-
gies that require medical, imaging, and video packages (Abdullah et al. 2013:
Cho et al. 2014: Fiss et al. 2015; Abdullah et al. 2012; Gortler et al. 1996; Greig
et al. 1989).

FIGURE 9.1
Binary graphic cutting value of brain cancer.
Performance Analysis of Machine Learning Algorithm 117

9.2 Literature Review
Segmentation is one of the most significant assignments in the field of PC
vision and has been studied for a long time. One of the best-known image
splitting techniques is grab-cut (Mendrik et al. 2015; Gustavo et al. 2018; Gupta
and Pahuja 2017), a directed frontal area/foundation division technique for
2D pictures that has just been executed in many picture altering program-
ming applications. Grab-cut depends on realistic cutting calculations (Saeed
and Abdullah 2019; Saeed and Jafri 2015; Horn and Chen 2007). Illustration
slices can be applied to dimensional information, including pictures, video
groupings, and 3D structures (Jarabo et al. 2014), and can likewise be reached
out to numerous segmentation labels (Kolmogorov and Zabih 2002). This
recommends a division strategy for pictures utilizing a 4D light field that
utilizations nearness and difference signals (Kowdle et al. 2012), as used for
regulated strategies. They train an arbitrary woods classifier to fuse appear-
ance and disparity so as to manage these various sorts of data in nature and
get a particular likelihood for each brand. Notwithstanding its prosperity,
the main way that division results happen is the information is viewed as
a schematic structure with vertices and edges in methods for cutting the
guide. The head is at the highest point of every pixel, and, in light of its com-
parability, the contiguous pixels are associated with a reasonable perspec-
tive. This mark likewise has an uncommon peak called a terminal in cases
including division of various blurb images. Pixel heads are joined to all sta-
tions, where the brand’s edge loads decide the probability of order. The best
approach to discover pieces in the diagram at the most minimal expense is
to get hash with minimal measure of accessible vitality and this issue will be
tackled utilizing the base greatest stream calculation (Levoy and Hanrahan
1996; Lin et al. 2015; Saeed et al. 2018), two-dimensional segmentation pic-
ture. Growing vertex is attached to one end after cutting, which means the
opposite poster is set to the opposite side. Our approach also uses the pro-
cess of splitting the scheme. Such methods of segmentation are known as
moderated methods because in the form of guides they require user input.
While some methods of graphics clipping (Saeed et al. 2019a, Maeno et al.
2013, and Marx 2009) can handle data from any distance, they are not always
ideal for high-dimensional data like video clips. Video data has an incon-
sistent structure along the time axis, unlike 3D storage data. Therefore,
methods of segmentation can be strengthened by taking proper account of
dysfunctional neighborhood relationships (Nagahashi et al. 2009). Enhance
the best of video segmentation by way of defining temporary neighborhood
relationships similar to neighborhood frames pixels. One trouble with clean
4D field data that is much like video facts is that iteration is obvious in a
complex field. Our working approach is the primary technique that makes
use of the image cut technique to concentrate on segmenting a 4D light field.
In the meantime, certain unsupervised processes can be used to segment
118 Machine Learning for Healthcare

multi-huge pictures or 4D spotlights. The author (Adelson and Bergen 1991)

Suggested a 4D light field segmentation primarily based on a level setting
technique (Ng et al. 2005), which applies an active contour technique to a
massive 4D volume phase. The author (Osher and Sethian 1988), suggest a
method for depth tagging multiple-width images based on the fact that fore-
ground objects cannot be excluded by deeper objects. The researcher sug-
gested a method to automatically extract objects from multi-width images
using the contrast signal. The researcher’s technique uses contrast and
appearance signals in multiple display images to determine the probability
of foreground objects (Platt 1999).To identify areas of transparent species as
a bright zone, an uncensored approach is proposed. The method makes use
of the light field distortion feature (Rother et al. 2004), which represents the
likelihood of a pixel belonging to an obvious object area, in addition to a seg-
mented binary photographs segmentation method. While those techniques
are a success, uncensored methods aren’t suitable for truly selecting the for
field editing due to the fact the regions of interest vary from user to person
(Wanner and Goldluecke 2012).
This proposes a division technique for pictures utilizing a 4D light field that
utilizations nearness and differentiation signals like (S. Wanner, Meister, and
Goldluecke 2013), as for regulated strategies. They train an irregular back-
woods classifier to fuse appearance and imbalance so as to manage these
various kinds of data in nature and acquire a particular likelihood for each
brand. In spite of its prosperity, the main way that division results happen is
the two-dimensional central images (Wanner, Straehle, and Goldluecke; Xu
et al. 2015; Saeed et al. 2019b).

9.3 Methodology
The basis of this study is the detection of brain cancer due to the interaction
of MRI-4D images with LFT segmentation. The researcher discusses the cel-
lular damage of brain cells or tissues due to brain cell abnormalities. The
main goal of our work is to build a framework that can recognize or isolate
the CSF spill field between tumors and not silent tumors. Initially, an MRI
image is equipped with a specific final target to shape the image for the rest
of the procedures.

9.3.1 Proposed Framework
The examination for the most part, comprises of two parts, which include:
Light Field Toolbox, high dimensional light field segmentation method,
high dimensional structured graphs
Performance Analysis of Machine Learning Algorithm 119

9.3.2 Light Field Toolbox for MATLAB

The purpose of using a set of tools to work with luminous images (also
known as plenoptic images) in MATLAB is to decode, calibrate correct, cor-
rect, color, filter basic images, and display luminous images.

9.3.3 High Dimensional Light Field Segmentation Method

The purpose of choosing a light field is basically a three-dimensional struc-
ture as the sounds of each pixel corresponding to the rays. The two dimen-
sions determine the location of those rays, while the other two determine their
direction. In the case of images measured by a camera based on the viewfinder
such as Lytro, the two dimensions choose a lens image. The other two choose
pixels within the image of this lens to maintain fractionation accuracy.

9.3.4 High Dimensional Structured Graphs

The purpose of choosing a high dimensional pattern structure is to imple-
ment the pattern cutting algorithm to reduce energy function. The solution
provides the optimum mark for every beam. It has been observed that the
solution becomes almost perfect for dividing multiple classifications.

9.4 High Dimension Structured Graphs

Grab-Cut is representing the improved method of light field image segmen-
tation process which is given below:

9.4.1 Grab-Cut
Grab-Cut is a graphic form of clip art. The algorithm estimates the color dis-
tribution of the target object and the background color using the Gaussian
Mix model and begins with a user-defined schema in the object into a section.
This is used to create a random Markov field in pixel tags using the control
function. The same poster is preferred in continuous areas and the improve-
ment of the graph whose values ​​can be inferred because this estimate may be
more accurate than the original taken from the bounding frame. This system
is repeated in two steps until convergence takes place. Incorrectly classified
and optimized place users also can accurately estimates. The process addi-
tionally corrects the effects to keep the edges.
Figure 9.2 shows the remaining area after the cutting of the whole image
as we can see that the above area is more clear and visible with colored form.
120 Machine Learning for Healthcare

FIGURE 9.2
Selecting the region of the whole image.

FIGURE 9.3
Identify the binary image of selecting region of image.

9.4.2 Image Testing Value

Figure 9.3 shows the remaining area after the cutting of the whole image
as we can see that the above area is more clear and visible after the binary
transformation of image segmentation process which converts into black
and white form that will create more visible to detect diseases.

9.4.3 Image Testing Result

Figure 9.2 shows the more visible area after the filtering tool implementation
that can appear clearer and more useful in identifying the diseases as well.
Performance Analysis of Machine Learning Algorithm 121

Figure 9.3 shows the more visible area after the filtering tool implementa-
tion that can appear more clear and useful to identify the diseases as well
with binary transformation as the above context shows that the result gener-
ates after the input of the second time. However, it is also is in the form of
black and white but we use BW filter for reefing the image more.

9.4.4 Graph Cut Value for B/W Image

Figure 9.4 shows the selected region of CSF matter inside the skull as the
researcher easily detects the selected damaged area of cancer. The above fig-
ure shows the selection area of input 1 for the segmentation process with
different pixel quality.
Figure 9.5 shows that the yellow color for cutting the edge of the area that
can easily and cut the specific area and transform into the next step of the
segmentation process. The above figure shows the selection area of input 2
for the segmentation process with a different resolution of the picture.

9.4.5 Image Testing Value

Figure 9.6 shows the more visible area after the filtering tool implementation
that can appear clearer and more useful in identifying the diseases as well
but the quality is less compare to color figures.

9.4.6 Image Testing Result

Figure 9.7 shows the more visible area after the filtering tool implementation
that can appear more clear and useful to identify the diseases as well with
binary transformation as the above context shows that the result generates
after the input of the second time.
As we can see the huge difference of both image type and after evalua-
tion, it is clear that the light field segmentation is better for color imaging

FIGURE 9.4
Graph cut value of original image.
122 Machine Learning for Healthcare

FIGURE 9.5
Graph cut value of input 1.

FIGURE 9.6
Graph cut value of input 2.

FIGURE 9.7
Original colors of image testing result for binary image.
Performance Analysis of Machine Learning Algorithm 123

transformation to compare to black and white image because after the imple-
mentation the quality of the image is more visible that can we prefer to use
color images in segmentation form.

9.5 Algorithm
The 4-dimesion algorithm is given below:

Algorithm.1: For Detection of CSF Spot in Image

(Two Outputs and One Input):

1.% Translate RGB image into L*a*b* color space.

2.X = rgb2lab(RGB);
3.% Create empty mask.
4.  BW = untruthful(size(X,1),size(X,2));
5.  % Draw Freehand
6.  m = size(BW, 1);
7.  n = size(BW, 2);
8. addedArea = poly2mask(xPos, yPos, m, n);
9. BW = BW | addedRegion;
10. % Create masked image.
11. CoveredImage = RGB;
​​12​.​  Cov​eredI​mage(​repma​t(~BW​,[1 1 3])) = 0;
13.End;

Algorithm.2: Detection of CSF Image 1 with Binary Gary

Scale and Original Image Detection:

1.Variables have been created in the base workstation.

2.Whos:
3. BW1 → 670x1269 → 850230 logical
4. BW2 → → 2798​40 lo​gical​
​583x4​80 ​ ​
5. Covered → Image1 → 670x1269x3 2550690 uint8
6. Covered → Image2 → 583x480x3 839520 uint8
7.Figure, imshow(BW1)
8.Figure, imshow(maskedImage1)

Algorithm.3: For Texture Feature Image Include (One

Input Two Output)

1. % Translate RGB image into L*a*b* color space.

2. X = rgb2lab(RGB);
3. % Produce empty front-size.
4. BW = untruthful(size(X,1),size(X,2));
124 Machine Learning for Healthcare

5. % Draw Freehand
6.   m = size(BW, 1);
7.   n = size(BW, 2);
8. AddedArea = poly2mask(xPos, yPos, m, n);
9. BW = BW | addedArea;
10.% GenerateCovered image.
11. CoveredImage = RGB;
12. Covered Image(repmat(~BW,[1 1 3])) = 0;
13.End;

Algorithm.4: For size of images

1.Variables have been shaped in the base workspace.

2.Whose
For Name, Size, Bytes Class and Attributes
3.BW1 → 670x1269;
4.BW2 → 670x1269;
5.maskedImage1 → 670x1269x3;
6.Covered Image2 → 670x1269x3;
7.Figure, imshow(BW2);
8.Figure, imshow(Covered Image2);

9.6 Result and Discussion

In this research article, the researcher reveals brain cancer researchers because
of the interface of MRI-4D images. Through the use of 4D image field seg-
mentation. Initially, MRI handled the pre-prepared image method with the
ultimate goal of adjusting the image for the rest of the procedure. Based on this
study, brain cancer and CSF were detected due to the interface of the 4D image
segmentation process. Therefore, it consists of a research study after primary
and secondary sources, followed by a cutting technique from the MATLAB
drawing tool using images from original medical samples to measure the
extent of damaged cells in the brain within the LCR. To implement the method
of cutting graphics using MATLAB tools and algorithms; in this study, the
researcher proposes a 4D modulation method that monitors the light field that
can be used to emit light with color shift and binary. By building a structured
4D diagram, the 4D light fields can be divided to reduce the graphics algo-
rithm generated by the 4D diagram. The researcher uses this technique to free
the damaged brain samples from the brain skull. These results demonstrate
the effectiveness of our approach to light editing applications. Lighting field
methods can be useful in improving the quality of photo editing application
and compound lighting field tubes, as they reduce border artistic effects.
Performance Analysis of Machine Learning Algorithm 125

The researcher evaluates ways to overcome the value of lost data in the
computational experiments of the proposed new method. We use a struc-
tured graphics break algorithm with correlation and multiple types of data
loss in time series data. The researcher explained that improving the cal-
culated data to determine the variables of the validation method is related
to the time delay of the test and the training vector resulting from the time
delay. The graphics cut method demonstrates the accuracy of the images
with the results and maintains the color shift with the color shift and binary
and produces one output and one input.

9.7 Conclusion
In this article, the researcher indicates a supervised 4D area for dependent
4D graphics, 4D light field pictures may be segmented with the aid of a pic-
ture clipping algorithm of rules. Experimental outcomes display that our
method achieves extra accuracy than previous methods using general light-
ing field data sets from the UK’s Brain Cancer Research Centre and CSF. In
addition, the researcher applied the proposed method to the original images
and showed the result in a color and binary shift with one and two outputs.
These results show the efficiency of our process of light field photo editing
using the algorithm of graphics. The problem is calculation time, as using a
graphic cutting algorithm requires large quantities of calculation time when
there are many peaks; the obvious future goal is to solve this issue.

9.8 Future Work
The features of the tumor are detected and will be useful for therapeutic use.
The researcher analyzes all results in the 2D and 3D sense, and also gener-
ates the novelty of MRI tools and Future work also includes segmenting and
identifying further images with more functionality helping to identify mul-
tiple types of tumors.

Acknowledgment
Authors are grateful to the Department of Software Engineering, faculty of
computing, Universiti Teknologi Malaysia-UTM, Malaysia for financial sup-
port to carry out this work.
126 Machine Learning for Healthcare

References
Abdullah, Afnizanfaizal, et al. “Cerebrospinal fluid pulsatile segmentation-a review.”
In: The 5th 2012 Biomedical Engineering International Conference, IEEE, 2012.
Abdullah, Afnizanfaizal, et al. “An improved swarm optimization for parameter
estimation and biological model selection.” PLoS One 8(4) (2013): 4.
Adelson, Edward H., and James R. Bergen. The Plenoptic Function and the Elements
of Early Vision. Vol. 2, Vision and Modeling Group, Media Laboratory,
Massachusetts Institute of Technology, 1–12,1991.
Berent, Jesse, and Pier Luigi Dragotti. “Unsupervised extraction of coherent regions
for image based rendering.” BMVC, 1–10,(2007).
Bishop, Tom E., and Paolo Favaro. “The light field camera: Extended depth of field,
aliasing, and superresolution.” IEEE Transactions on Pattern Analysis and Machine
Intelligence 34(5) (2011): 972–986.
Boykov, Yuri, and Gareth Funka-Lea. “Graph cuts and efficient ND image segmenta-
tion.” International Journal of Computer Vision 70(2) (2006): 109–131.
Boykov, Yuri Y., and M.-P. Jolly. “Interactive graph cuts for optimal boundary &
region segmentation of objects in ND images.” In: Proceedings of the Eighth IEEE
International Conference on Computer Vision. ICCV 2001. Vol. 1, IEEE, 2001.
Boykov, Yuri, and Vladimir Kolmogorov. “An experimental comparison of min-cut/
max-flow algorithms for energy minimization in vision.” IEEE Transactions on
Pattern Analysis and Machine Intelligence 26(9) (2004): 1124–1137.
Boykov, Yuri, Olga Veksler, and Ramin Zabih. “Fast approximate energy minimiza-
tion via graph cuts.” IEEE Transactions on Pattern Analysis and Machine Intelligence
23(11) (2001): 1222–1239.
Chen, Can, et al. “Light field stereo matching using bilateral statistics of surface
cameras.” In: Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, 2014.
Cho, Donghyeon, Sunyeong Kim, and Yu-Wing Tai. “Consistent matting for light
field images.” In: European Conference on Computer Vision, Springer, Cham, 2014.
Fiss, Juliet, Brian Curless, and Rick Szeliski. “Light field layer matting.” In: Proceedings
of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
Gortler, Steven J., et al. “The lumigraph.” In: Proceedings of the 23rd Annual Conference
on Computer Graphics and Interactive Techniques, 1996.
Greig, Dorothy M., Bruce T. Porteous, and Allan H. Seheult. “Exact maximum a pos-
teriori estimation for binary images.” Journal of the Royal Statistical Society: Series
B (Methodological) 51(2) (1989): 271–279.
Gupta, Anjali, and Gunjan Pahuja. “Hybrid clustering and boundary value refine-
ment for tumor segmentation using brain MRI.” In: IOP Conference Series
Materials Science and Engineering, Vol. 225, no. 1, IOP Publishing, 2017.
Horn, Daniel Reiter, and Billy Chen. “Lightshop: Interactive light field manipulation
and rendering.” In: Proceedings of the 2007 Symposium on Interactive 3D Graphics
and Games, 2007.
Jarabo, Adrian, et al. “How do people edit light fields?” ACM Transactions on Graphics
(Proc. SIGGRAPH) 33(4):1–11, (2014).
Kolmogorov, Vladimir, and Ramin Zabih. “Multi-camera scene reconstruction
via graph cuts.” In: European Conference on Computer Vision, Springer, Berlin,
Heidelberg, 2002.
Performance Analysis of Machine Learning Algorithm 127

Kowdle, Adarsh, Sudipta N. Sinha, and Richard Szeliski. “Multiple view object
cosegmentation using appearance and stereo cues.” In: European Conference on
Computer Vision, Springer, Berlin, Heidelberg, 2012.
Levoy, Marc, and Pat Hanrahan. “Light field rendering.” In: Proceedings of the 23rd
Annual Conference on Computer Graphics and Interactive Techniques, 1996.
Lin, Haiting, et al. “Depth recovery from light field using focal stack symmetry.” In:
Proceedings of the IEEE International Conference on Computer Vision, 2015.
Lumsdaine, Andrew, and Todor Georgiev. “The focused plenoptic camera.” In: 2009
IEEE International Conference on Computational Photography (ICCP), IEEE, 2009.
Maeno, Kazuki, et al. “Light field distortion feature for transparent object recog-
nition.” In: Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, 2013.
Mendrik, Adriënne M., et al. “MRBrainS challenge: Online evaluation framework
for brain image segmentation in 3T MRI scans.” Computational Intelligence and
Neuroscience 2015 (2015).
Nagahashi, Tomoyuki, Hironobu Fujiyoshi, and Takeo Kanade. “Video segmenta-
tion using iterated graph cuts based on spatio-temporal volumes.” In: Asian
Conference on Computer Vision, Springer, Berlin, Heidelberg, 2009.
Ng, Ren, et al. “Light field photography with a hand-held plenoptic camera.” Computer
Science Technical Report CSTR 2(11) (2005): 1–11.
Oliveira, Gustavo Casagrande, Renato Varoto, and Alberto Cliquet Jr. “Brain tumor
segmentation in magnetic resonance images using genetic algorithm cluster-
ing and AdaBoost CLASSIFIER.” BIOIMAGIN, 2, 77–82 (2018).
Osher, Stanley, and James A. Sethian. “Fronts propagating with curvature-depen-
dent speed: Algorithms based on Hamilton-Jacobi formulations.” Journal of
Computational Physics 79(1) (1988): 12–49.
Platt, John. “Probabilistic outputs for support vector machines and comparisons
to regularized likelihood methods.” Advances in Large Margin Classifiers 10(3)
(1999): 61–74.
Rother, Carsten, Vladimir Kolmogorov, and Andrew Blake. “’rabCut’ interactive
foreground extraction using iterated graph cuts.” ACM Transactions on Graphics
(TOG) 23(3) (2004): 309–314.
Saeed, Soobia, and Afnizanfaizal Bin Abdullah. “Investigation of a brain cancer with
interfacing of 3-dimensional image processing.” In: 2019 International Conference
on Information Science and Communication Technology (ICISCT), IEEE, 2019.
Saeed, Soobia, and Afnizanfaizal Abdullah. “Recognition of brain cancer and cere-
brospinal fluid due to the usage of different MRI image by utilizing support vec-
tor machine.” Bulletin of Electrical Engineering and Informatics 9(2) (2020): 619–625.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi. “Analysis of the Lung
Cancer patient’s for Data Mining Tool.” IJCSNS 19(7) (2019a): 90.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi. “Implementation of
Fourier transformation with brain cancer and CSF images.” Indian Journal of
Science and Technology 12(37) (2019b): 37.
Saeed, Soobia, and Raza Jafri. “Estimation of brain tumor using latest technology of
mobile phone.” Journal of Information and Communication Technology (JICT) 9(1)
(2015): 32–09.
Wanner, Sven, and Bastian Goldluecke. “Globally consistent depth labeling of 4D
light fields.” In: 2012 IEEE Conference on Computer Vision and Pattern Recognition,
IEEE, 2012.
128 Machine Learning for Healthcare

Wanner, Sven, Stephan Meister, and Bastian Goldluecke “Datasets and benchmarks
for densely sampled 4D light fields.” VMV 13, 1–12, 2013.
Wanner, Sven, Christoph Straehle, and Bastian Goldluecke. “Globally consistent
multi-label assignment on the ray space of 4d light fields.” In: Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition, 1–12,2013.
Xu, Yichao, et al. “Transcut: Transparent object segmentation from a light-field
image.” In: Proceedings of the IEEE International Conference on Computer Vision,
1–8, 2015.
10
Patient Report Analysis for Identification
and Diagnosis of Disease

Muralidharan C., Mohamed Sirajudeen Y., and Anitha R.

CONTENTS
10.1 Introduction................................................................................................. 130
10.2 Data Variability........................................................................................... 131
10.2.1 Structured Data............................................................................... 132
10.2.1.1 Human Generated Data.................................................. 132
10.2.1.2 Machine Generated Data................................................ 132
10.2.2 Semi-Structured Data .................................................................... 133
10.2.3 Unstructured Data ......................................................................... 133
10.2.4 Comparison of Structured, Unstructured Data, and
Semi-Structured ............................................................................. 134
10.3 Data Collection of Diseases ...................................................................... 135
10.3.1 EMR Data Collection through eHealth Devices ........................ 135
10.3.2 Semantic Data Extraction from Healthcare Websites................ 136
10.3.3 Patient Chatbots.............................................................................. 136
10.3.4 Structured Data............................................................................... 137
10.3.5 Consistency and Quality of Structured Data ............................ 137
10.4 Predictive Models for Analysis................................................................. 138
10.4.1 Regression Techniques .................................................................. 139
10.4.2 Machine Learning Techniques..................................................... 140
10.4.3 Algorithms....................................................................................... 141
10.4.3.1 Naïve Bayes....................................................................... 141
10.4.3.2 Support Vector Machine................................................. 142
10.4.3.3 Logistic Regression.......................................................... 142
10.4.3.4 Decision Trees .................................................................. 142
10.4.4 Use Cases......................................................................................... 142
10.4.4.1 Cleveland Clinic............................................................... 142
10.4.4.2 Providence Health ........................................................... 143
10.4.4.3 Dartmouth Hitchcock ..................................................... 143
10.4.4.4 Google................................................................................ 143
10.5 Semi-Structured Data................................................................................. 144
10.5.1 Semantic Extraction........................................................................ 144
10.5.2 Web Mantic Extraction................................................................... 144
10.5.3 Use Cases......................................................................................... 145

129
130 Machine Learning for Healthcare

10.6 Unstructured Data...................................................................................... 145

10.6.1 Finding Meaning in Unstructured Data .................................... 145
10.6.2 Extraction of Data........................................................................... 146
10.6.2.1 Text Extraction.................................................................. 146
10.6.2.2 Image Extraction.............................................................. 146
10.6.2.3 Challenges of Data Extraction from PDFs.................... 147
10.6.2.4 Video Extraction............................................................... 152
10.6.2.5 Sound Extraction.............................................................. 153
10.6.3 Algorithms....................................................................................... 153
10.6.3.1 Natural Language Processing........................................ 153
10.6.3.2 Naïve Bayes....................................................................... 154
10.6.3.3 Deep Learning.................................................................. 154
10.6.3.4 Convolutional Neural Network..................................... 154
10.6.3.5 Phenotyping Algorithms ............................................... 155
10.6.4 Use Cases......................................................................................... 155
10.7 Conclusion .................................................................................................. 156
References.............................................................................................................. 156

10.1 Introduction
Healthcare is an important field where the health of living beings is improved
or maintained through diagnosis, prevention, and treatment of diseases. The
costs for the healthcare have been increasing constantly and on the other
hand, the quality of care given to the patients has seen considerable improve-
ments. Several researchers in health care industry have come up with a study
which shows that after the incorporation of current healthcare technolo-
gies, the mortality rate, cost, and complications have been reduced a con-
siderable amount.In 2009, the US government created a Health Information
Technology for Economic and Clinical Health Act (HITECH) which includes
an incentive scheme worth around 27 billion US dollars for use of Electronic
Medical Records. The advances in the IT industry include the ability to col-
lect healthcare data that exists in various forms. Data seems to be an integral
part of the healthcare field. A report by Google comments on big data (Luo
et al. 2016) indicates that existing healthcare data has the potential worth of
300 billion US dollars because of the advancements in the technologies that
enables the sensing and acquisition of data. Healthcare institutions or orga-
nizations and hospitals are collecting patients’ healthcare data. Advanced
analytical techniques need to be developed for better understanding and
knowledge of the healthcare data, as it may transform the existing data to
meaningful information (Sharmila et al. 2017). Data analysis forms a critical
component of these emerging computing technologies. The solutions that
are observed from the analysis are then applied to the available healthcare
Patient Report Analysis  131

data which have the potential for transforming healthcare from a reactive
state to a proactive state (Mohan and Sarojadevi 2018). It has been predicted
that for several years healthcare analysis will grow more and more. Typically,
the underlying patterns of several diseases can be observed and understood
by analyzing health data. This allows the physicians to build a personalized
patient profiles which can support the physician for computing accurate diag-
noses for the individual patients who are likely to suffer diseases. Healthcare
data are the most valuable data and can be derived from a variety of sources
such as sensors, clinical notes, images, text from biomedical literatures, or
traditional electronic records (Jutel 2011). These different types of data are
collected from different sources seem to be heterogeneous in nature, which
requires several challenging processes for analysis. Various techniques are
needed for analyzing the different forms of data. Due to the heterogeneous
nature the data integration seems more challenging. In many of the cases
the insights are obtained from diverse data types that cannot be collected
from single source. Hence, high potential integrated data analysis methods
are needed (Razia et al. 2017). The healthcare field observes many advances
that are coming from diverse disciplines such as data mining, databases,
information retrieval, healthcare practitioners, and medical researchers.
This booming interdisciplinary nature adds richness to the healthcare field,
but it also adds challenges for making significant advances. This is because
researchers from the computer science field will not have previous exposure
to domain-specific medical concepts. Likewise, practitioners and researchers
of medical fields will have had limited exposure to the statistical and math-
ematical concepts that are required for data analytics (Vinitha et al. 2018).
This understood to be a critical situation for creating the coherent body of
work in this field, even though the available analysis techniques can process
the available data. This diversity results in forming an independent line of
work that is based on two completely different perspectives. As a result, this
chapter aims to provide better understanding about healthcare data and its
different forms with some possible algorithms and uses cases.
The chapter has been organized so that Section 10.2 examines the data
variability; Section 10.3 focuses on data collection; Section 10.4 describes the
structured data; Section 10.5 describes the semi-structured data; Section 10.6
briefly explains the unstructured data; and Section 10.7 concludes the chapter.

10.2 Data Variability
Put simply, data is something that provides valuable information by analyz-
ing it. Data are of different sizes and formats. For example, data information
that are provided in a resume will be about a particular person and will
include educational details, working experience, personal interests, address,
132 Machine Learning for Healthcare

etc. Small sized data can be easily analyzed or understood. But in recent
times, the size of digital data has increased tremendously. This may include
different forms of data that need to be collected and analyzed for extracting
information. Digital data can come in different forms such as structured,
unstructured, and semi-structured.

10.2.1 Structured Data
Structured data are organized or labeled data that can be analyzed easily
and effectively. They are well organized data that are formatted in a reposi-
tory such as a database. This includes all the data that are stored in an SQL
database as rows and columns. Usually, a database has relational keys with
which the fields can be easily mapped, hence accessibility and searching
information is too easy in such type. They are comparatively too simple for
storing, retrieving, and analyzing, but are strictly defined in terms of field
type and field name. Nowadays, these types of data seem to the be most pro-
cessed, as it is simpler for information processing, but they represent only
10% of all the informatics data. Thus, it is not be sufficient to extract the infor-
mation by analyzing the data. An example is relational data.
The data source can be from two different sources such as:

1. Data generated by humans

2. Data generated by machines

10.2.1.1 Human Generated Data

These data are generated by humans by making an interaction with machines.
Examples include:

Input data: Input data are fed to a machine by a human. For exam-
ple, to understand customer behavior data like their name, sex, age,
income, and survey responses (which are non-free) and so on will
be collected.
Click stream data: This data can be generated from websites when-
ever a link is clicked. This might be analyzed for acquiring valuable
information.
Gaming-related data: In games every move made by the gamer will be
recorded for understanding the behaviors of the different users so
that the game can be updated.

10.2.1.2 Machine Generated Data

These data are generated automatically without any interaction by the
human.
Patient Report Analysis  133

Examples include:

Sensor data: this includes RFID tags, healthcare devices, smart meters,
GPS data, etc. These can be used in the inventory control and supply
chain management.
Log data: the behavior of the applications, servers, and networks will
be recorded every now and then while they operate. These behav-
ioral logs are called log data. These logs will be in huge forms which
can be used to predict the security breaches that occur and other
changes in the service level agreements.
Point-of-sale data: the product related information can be generated
when the bar code of the product is scanned while it is purchased.
Financial data: systems used in financial sectors are automated
through predefined rules. For example, trading data which contains
the symbol of the company and its dollar value.

10.2.2 Semi-Structured Data
Semi-structured data are structured data formed in an unorganized way.
This type of data has the properties related to an organization but will not
reside in the relational database and can be processed easily. By making
changes in the process, it can be stored in the relational database. Since it does
not have the formal structure as a relational database or any other form of
data tables, it needs tags or other form of markers for separating the semantic
elements, hierarchies, and fields that exist within the data. Examples include
the data in websites such as JSON files, .csv files, XML files, and delimited
text files. Since this type of data are in unorganized forms it is difficult to
store, retrieve, and analyze.

10.2.3 Unstructured Data
The data that is in unorganized formats or that do not have data models are
said to be unstructured data. Therefore, it will not fit the relational database
in a predefined manner. Thus, for these types of data, alternative advanced
tools (software, etc.) will be used for storing, accessing, and managing the
data. This type will be most prevalent in IT systems for varieties of business
intelligence, and for different analytics applications for predicting valuable
information. Examples include Word files, PDFs, images, videos, audio, text,
web pages, emails, and other streaming data.
Examples of unstructured data (generated by machines) include:

• Satellite data: data collected from satellites such as surveillance

imagery, weather, etc.
• Scientific data: atmospheric data which are collected by sensors and
machines.
134 Machine Learning for Healthcare

• Photographs and video: surveillance footage, traffic monitoring

videos, etc.

Examples of unstructured data (generated by humans):

1. The internal text of enterprises: this includes documents, logs, and

e-mails that are maintained for an organization. The information of
the enterprise seems to generate the largest text information.
2. Social media data: data generated from social media outlets such as
Youtube, LinkedIn, Flickr, etc.
3. Mobile data: data collected from mobile devices include location
information, text messages, user logs, etc.
4. Website content: the data that are collected from any website will be
in unstructured format. This may include data gather from Twitter,
and Instagram, etc.

10.2.4 Comparison of Structured, Unstructured

Data, and Semi-Structured
Table 10.1

TABLE 10.1
Comparison of Structured, Semi-Structured, and Unstructured Data
STRUCTURED SEMI-STRUCTURED UNSTRUCTURED
Field DATA DATA DATA
Technology Relational database XML/RDF Binary data
Management of Versioning can be Versioning is possible Versioning as the
Versions done over rows, through graph or tuples whole
tuples, tables.
Management of Transaction is Transaction is based on No concurrency and
Transactions matured and uses DBMS and are not transaction
concurrency matured. management
Techniques
Scalability of Low Medium High
data
Flexibility of Flexibility is low Flexibility is higher than More Flexible and no
data and is schema structured data and lower schema dependency
dependent than unstructured data
Robustness Highly robust Not very spread as it is —
new technology
Performance of Queries are Anonymous nodes can be Query over text is
Query Structured which queried alone possible
allows joining of
complex data
Patient Report Analysis  135

10.3 Data Collection of Diseases

Data collection is defined as the process of collecting, analyzing, and inter-
preting different types of information relating to a particular disease or
healthcare need. Traditional patient records are collected from sources like
personal surveys, handwritten prescriptions, and hardcopies of the patient’s
records from local hospitals. Prior to the evolvement of digital data, the
healthcare records come in physical form. Thus, the data are collected and
managed within the hospital itself. But following the recent IT advance-
ments, patient records are collected in a digital format (Kaur and Siri 2006).
Some of the examples of digital data used in the field of medicine are digi-
tal scan reports, videos shot on laparoscopic cameras, digital X-ray reports,
endoscopy videos, and ultrasonic records. These medical data are fast grow-
ing data in the digital world. As per the survey conducted by DELL EMC.
(2018) the healthcare data growth rate has increased by 878% since 2016. It
also claims that the total amount of healthcare data will have reached 20,000
petabytes by 2020. In addition to that, more healthcare applications and data-
bases are developed every day that work with healthcare data.
An important source of electronic medical records are electronic health
(eHealth) devices and communication-supported health devices. These col-
lect data at frequent intervals from patients though eHealth devices and store
it in cloud storage. If the data are collected from patients through electronic
devices directly, then the data are called Patient-Generated Healthcare Data
(PGHD). The Cloud Service Provider (CSP) maintains the patient’s clinical
data like demographics, progress notes, problems, and medications on cloud
storage. Patients’ medical records are digitalized and assist in ensuring data
is accurate. Electronic Medical Records (EMR) data collection can be clas-
sified as both quantitative and qualitative data collection. In quantitative
data collection, the data are collected in the form of numeric variables. In
other words, the information is collected from the patient as numeric values,
such as count, number, and percentage. Qualitative data collection methods
collect patient data in a non-numeric fashion. This type of data is collected
through methods of observation, one-to-one interviews, and online surveys.
Qualitative data are also known as categorical data.
The important ways of collecting the Electronic Medical Records (EMR)
are eHealth devices, semantic data collection, and patient chatbots.

10.3.1 EMR Data Collection through eHealth Devices

eHealth devices are also called self-monitoring healthcare devices. They use
sensors and wireless communication design to measure the patient’s health
and transfer it to cloud storage. This allows the patient, as well as the phy-
sicians, to measure and monitor the patient’s health remotely. Some of the
136 Machine Learning for Healthcare

available healthcare monitoring eHealth devices in the market are tempera-

ture devices, heartbeat tracking devices, glucometers, oximeters, pulsometers,
and blood pressure devices. These IoT-based healthcare devices are consid-
ered to be an important advancement in the field of healthcare management.
As the use of cloud computing and wireless technology increases, the demand
for eHealth devices is also rapidly growing. It is predicted that in 2020, eHealth
devices will account for 80% of wireless devices. The main advantage of these
devices are mobility and accessibility of smartphones and tablets.

10.3.2 Semantic Data Extraction from Healthcare Websites

Semantic extraction of healthcare information extracts information related
to a particular disease, medical facts, attributes from a website, or unstruc-
tured data. The purpose of semantic data extraction in healthcare is to enable
analysis of the unstructured content, electronic prescriptions, medical text
documents, emails, digital images, and patient reports. The main objective
of semantic analysis is to structure the unstructured data (Wu et al. 2018).
Semantic data extraction on websites has two major approaches: rule
matching data collection and machine learning data collection

Rule matching data collection: this collects the information related to

a particular word or phrase from websites. A rule-based matching
algorithm is used on raw medical websites to gather the information
about a particular disease. They also provide access to the tokens
within the document and their relationships.
Machine learning–based data collection: this is a statistical analysis
of the content, the potential compute-intensive application that can
benefit from using Hadoop. This approach derives the relationship
from statistical co-occurrence within the website.

10.3.3 Patient Chatbots
To deliver quality services to the patients, medical informatics entities are
using recent technologies like Artificial Intelligence and predictive technolo-
gies in the healthcare application. It is impossible for a patient to get advice
from physicians in an emergency situation. To provide “round the clock”
medical advice to patients, healthcare industries are investing a lot in the
creation of automated medical chatbots. Medical chatbots are conversational
software available for smartphone applications. They provide a more imme-
diate service for patients. They are adequate enough to communicate and
gather information from the patients. The collected information is fed to the
deep learning algorithms to improve the intelligence of the chatbots. These
medical chatbots are a recent trend in the healthcare industry. Some of the
most popular chatbots are related to the healthcare industry.
Patient Report Analysis  137

10.3.4 Structured Data
Medical data exists in different forms such as laboratory test results, notes by
physicians, lifestyle data of patients, vital signs, and various forms of imag-
ery data such as Magnetic Resonance Imaging (MRI), radiology, ultrasonog-
raphy, pathology slides. etc. There is no proper standard for encompassing
the medical data, hence it is important to understand the information of the
data before processing it.
Structured data are organized and consistent in nature. Structured data
can be analyzed easily. A few examples of medical data include numerical
values such as blood pressure, height and weight, and categorical values,
such as blood type, diagnostic stage of disease, etc. It is a non-homogenous
and non-monolithic category as the data will be in structured form and it
doesn’t mean that it makes sense with the data as it is in structured form.
Furthermore, we cannot say that the data with no formal structure cannot be
interrupted easily.

10.3.5 Consistency and Quality of Structured Data

The structured data consists of two main parts – the value and the vari-
able name. Consider the height of patients. In electronic medical records
of patients, the height of the patient might be stored as “height: 64”. This
depicts the height of the patient in inches. It is also possible to store the
value in meters as “height: 1.625m” or it might be stored in terms of yards
as “1.77 yards” and so on. The variable might also be stored in different
forms.
Logical Observation Identifiers Names and Codes (LOINC) is a univer-
sal and database standard developed for identifying the medical laboratory
reports. Health Level 7 is an international standard for transferring admin-
istrative and clinical data between the application which are used by various
medical providers. The Fast Healthcare Interoperability Resources (FHIR), is
a drafting standard for elements, data formats, and API for exchanging the
medical data.
Usually structured patient generated data is collected from devices that
are held with the consumers and it may not be an FDA approved device,
hence the data from these devices cannot be compared to each other
although they are in structured format. For example, when using an accel-
erometer the number of steps walked by a consumer is measured but there
is no standard algorithm for converting this raw data. Though this data
is inconsistent, the clinician still uses it to find the relative improvement
of patient for a period of time, conditionally the patient should use the
same electronic device for the entire period. But the direct patient compari-
son wiould be implausible. To overcome these issues standardized devices
should be used.
138 Machine Learning for Healthcare

10.4 Predictive Models for Analysis

The predictive models that exist in the data analytics provide a valuable
score for measurable medical data elements. These can also predict and pro-
vide the probability of diseases that can affect the patient in the near future.
It may also provide the likelihood of a patient defaulting on a disease based
on his or her personal history or characteristics. Statistical models are also
available for predicting the type of disease and prevention of diseases.
The context of applying predictive analytics is quite diverse. The expected
outcomes may vary from binary values such as “yes/no” or “true/false” for
fake prediction to predict the real numerical values of medical field (Asif
et al. 2018). Here, different classes of predictive techniques have been dis-
cussed which will support the reader to understand the various models that
are prevalent.

1. Managing the Data Sources: for managing the data sources, the
quality and operational pathways need to be understood. The
data may come from different sources such as a staging platform,
warehouse, or a “data mart” which is owned by an application.
The most important thing for the production is change manage-
ment. This is because while upstreaming the data sources the
change tolerance or anticipation and is important and must not
be restricted by the coming changes. To get into the main data,
it is suggested to have an engine related to the medical field for
the parsing and use of the structured data in the form of Extract,
Transform, and Load (ETL).
2. Metadata Management: data about the data is metadata. An example
can be seen in that the codes of ICD-10 will have different meanings,
but with these diagnostic codes the risk cannot be well understood
anf it will not be sufficient. The biggest risks are the most expen-
sive diseases; dangerous patient health conditions which come from
the concomitant interactions and an understanding of the disease
relationship is almost a prediction. For organization which are not
capable of understanding these relationships, specifications such as
the CMS-DRG codes can be used.
3. Population Segmentation: almost all the medicare predictive ana-
lytics need to create the population segments. Even to evaluate the
goal of the provider, having an understanding of the populations is
important. Thus, population segments are one of the core capabili-
ties for healthcare analytics. An important element is the physiologi-
cal state of the patients and this is known from the primary data
source. Segmenting the population is also important for responding
to interventions.
Patient Report Analysis  139

4. Data Manipulation: using most of the ETL functions, the data can
be processed, transformed, compared, and managed with more
aggressive capabilities for natural language processing. The tools
for manipulation are available as free resources, and the terminol-
ogy engines seem to be more sophisticated as these capabilities are
embedded within them.
5. Predicting Healthcare Facts: this is the core capability of the whole
process. It is important that the solution from the prediction sup-
ports the system to act upon the human health. Integrating the pre-
dictions with the systems that are exposed to customers must be
more simplified for convenience (Chen et al. 2017).

Predictive analytics can be classified into two major learning methods: super-
vised and unsupervised. In supervised learning the target variable is easily
acquired using traditional techniques. The trained data can be used to create
a model that can reveal the correlations between target and input variable.
Whereas in unsupervised learning, patterns and trends in the data need to
be discovered without any pre-defined assumptions. Attribute-based algo-
rithms can be constructed through this learning process by determining the
association and clusters in the data.
Predictive analytics can be grouped into two major groups: regression and
machine learning techniques.

10.4.1 Regression Techniques
Regression analysis is one of the predictive modeling techniques which pre-
dict the dependency between the target variables. It is used to predict sales
trends, and the possibility of churns or fraudulent transactions. It focuses on
forming the mathematical equation for capturing the interactions between
the different targets, thereby reducing the overall error in the predicted
model
Linear regression is one type of regression model that can be used to pre-
dict the response variable in a linear manner. The parameters can be adjusted
or learned so that the addition of squared residuals is minimized.
Logistic regression assigns probabilities for the possible outcomes. A
binary outcome variable can be converted to an unbounded continuous vari-
able from which a regular multivariate model is estimated.
Time series models are used for predicting the future behavior of the vari-
ables when the internal structures such as trends, auto correlation, etc. is
available. They are capable of analysing the components such as seasonal
and trends through which better models can be produced. A few time series
models include: the Moving Average Model, the Auto Regressive Model.
A combination of these two models is called an Auto Regressive Moving
Average and Auto Regressive Integrated Moving Average.
140 Machine Learning for Healthcare

Decision trees is a model which sees the collection of defined rules based
on variables in the dataset, where the rules are defined so as to obtain the
best split for differentiating the observations that belong to different target
classes. Rules are explanatory and are preferred by the data analysts.

10.4.2 Machine Learning Techniques

Machine learning–based models are another form of predictive analytics that
are used for applications such as diagnosing medical conditions, or fraud
detection, and more. However, unlike classification or regression trees, this
model remains a black box without considering the relationship between the
predictor variables, and it sufficiently predicts the dependent variable.
Among the various existing machine learning models, neural networks
are the model which is inspired by the human nervous system. It has become
increasingly popularity in recent times as it is capable of learning complex
relationships among the predictor variables. For classification there exists a
wide variety of neural network models. The earlier neural network models
used only three layers – the input, hidden, and output layer – and the deep
neural model has gained popularity by using more than one hidden layer.
Large numbers of neurons and the interconnections between them are capa-
ble of modeling non-linear relationships between input and output variables.
Some of the most commonly used neural networks that are suitable for
prediction tasks are:

1. Multilayer Perceptron
This neural network uses more than one hidden layer of neurons. It
is also known as “deep feed forward neural networks”.
2. Convolutional Neural Networks
This type of neural network performs convolutions between the
input data and desired filter. They are more efficient in learning
hierarchal features from the data by extracting the relationships
between the neighbors.
3. Recurrent Neural Network
These types of neural network have hidden layer neurons which
have self-connections for helping the neuron to possess memory.
These types of networks are suitable for text processing as the inter-
pretation of the text will be dependent on neighboring words or
contexts. Thus, these types of neural network models the interrela-
tionships of words by considering their sequence.
4. Long-Short Term Memory Network
This type of network are extensions of recurrent neural networks in
which each hidden layer neuron will be incorporated with a memory
cell. They are good at finding long-distance relationships. These types
of networks can be applied for analyzing any kind of sequential data.
Patient Report Analysis  141

10.4.3 Algorithms
Figure 10.1 shows the algorithms that are used for analyzing the healthcare
data. The algorithms used for analyzing structured data are discussed below.

10.4.3.1 Naïve Bayes
Naïve Bayes is a classifier method used for categorization of text, and for
solving the problem of document judging so as to identify the category to
which it belongs to. The Naïve Bayes classifier considers that one particu-
lar feature of the class will be unrelated to other features. Even though the
features of a class are independent, all its properties will independently con-
tribute its probability for a certain category. It is one of the most efficient clas-
sification algorithms that has been successfully applied for many medical
related problems.

FIGURE 10.1
Algorithms used for health care analysis.
142 Machine Learning for Healthcare

10.4.3.2 Support Vector Machine

Support Vector Machines (SVM) can be applied for both classification and
regression, but this algorithm is widely used in classification-related prob-
lems which divides the dataset into two different classes through a hyper-
plane. The goal of using the hyperplane is to find the greatest distance
between the hyperplane or possible margin at any point within the training
data, through which the new data can be correctly classified. In SVM, the
solution will always be the global optimum as it deals with the convex opti-
mization problems. They are extensively used in identifying cancer, cancer
prediction and diagnosis, detection of neurological diseases, and in general
for classifying datasets with imbalanced values.

10.4.3.3 Logistic Regression
Logistic regression is one of the multivariable algorithms which is used for
analyzing dichotomous outcomes. The procedure of logistic regression is
nearly the same as that of multiple linear regression, but the only differ-
ence is that the response variable will be binomial. When compared to linear
regression, the confounding effects are avoided by analyzing the variables’
association. In the healthcare field, it is widely used to solve classification-
related problems and for predicting the event probability.

10.4.3.4 Decision Trees
Decision trees are mostly used in the diagnosis of illnesses in the medical
field. In some cases, constant monitoring of auto neuropathy is required for
diagnosis, and sensors are used for the constant collection of medical data
from the patient for identifying the patterns in the datasets and for process-
ing of this data using machine learning algorithms. Identification of cardio-
vascular autonomic neuropathy is identified using the data from the sensors
by understanding the signs of diabetes. This analysis further supports the
development of advanced diet and treatment plans for the patient.

10.4.4 Use Cases
10.4.4.1 Cleveland Clinic
Cleveland Clinic aims to evaluate the quality of providers by creating risk
adjustment score. For evaluating the quality of the care delivered, it is impos-
sible to create an “apples to apples” comparison. To solve this problem, a
risk adjustment score is created for all the individuals who have suffered
with similar but unique situations, this allows the statistical models to be
predictive. Using sparsely annotated procedure codes, the Cleveland Clinic
compares the factors that are not related to the patient’s physiology using
Patient Report Analysis  143

the annotated procedure codes. This study has the ability to compare the
employers, health plans, and institutions while also accounting for the rep-
resented populations. This innovation uses simple data that can be availed
of from administrative records.

10.4.4.2 Providence Health
Providence Health demonstrates the management of high-risk patients by
collaborating multidisciplinary values. This disease care management pro-
gram can be used in relation to heart failure, diabetes management, asthma,
coronary disease, and COPD. It assesses the risk of the patient based on
type of emergency. Care teams collaborated with data scientists and felt that
working with simple and elegant solutions is more sufficient, as more com-
plex data mining is not possible.

10.4.4.3 Dartmouth Hitchcock
Dartmouth Hitchcock predicts readmission risk. It includes the creation of
readmission predictive models which are culled from Epic Clarity’s data
warehouse, and it has been found that Dartmouth remains one of the 22%
of American hospitals that avoids readmission penalties which are levied
by the Centers for Medicare and Medicaid Services. For congestive heart
failure, pneumonia, heart attacks, lung ailments such as chronic bronchi-
tis, and knee or hip replacements, the avoidance of unnecessary readmis-
sions the Dartmouth is in the top 2% of hospitals. This has shown that
Dartmouth is a place which ensures that patients are healing after they
leave the hospital.

10.4.4.4 Google
Google uses unlikely data sources to predict where the locations of flu and
dengue, which is semi-real time in nature. The team of data scientists col-
laborate with the epidemiologists from all over the world to parse the search
streams in order to predict, in real time, the whereabouts of flu and dengue.
In the case of the flu, they used surveillance data to validate the findings. In
the case of dengue, the epidemiological data from many developing coun-
tries are collected via Google. Based on the search stream, Google predicts
the presence of flu, and after few months the team validates the predictions
through further surveillance methods. In this case, a fundamental epidemio-
logical breakthrough took place which reduces the wait time as the disease
trend is achieved. In few years, this type of outside-the-box thinking will
be determining the extents to which healthcare systems dominate in a new
place, where the predictions will be embedded into administrative and clini-
cal software.
144 Machine Learning for Healthcare

10.5 Semi-Structured Data
Semi-structured data comes in the form of structured data which does not
rely on the formal structure of data models that are associated with rela-
tional databases or any other forms of data tables. It includes tags or mark-
ers for separating semantic elements. It enforces the hierarchy of fields and
records that exist within the data. This can also be called the self-describing
data model. Here, the entities which belong to single class will have different
attributes. Another important fact is that the attribute order is not important.
These types of data seem to increasingly occur since the emergence of the
internet. Semi-structured data are often found in object-oriented databases.

10.5.1 Semantic Extraction
Semantic extraction is a processing technique that is used to extract and
identify entities such as organizations, animals, locations, etc. for populating
meta data. The purpose of this model is to analyze unstructured data such
as text documents, images, emails, reports, and other business content. The
semantic analysis of unstructured data seems to be an important technique
as the unstructured data can be converted into structured data. The accuracy
of this model seems to be critical, because without a considerable level of
accuracy there is the risk of feeding the decision makers with non-actionable
or misleading insights.
Semantic extraction is based on one of two approaches:

• Rule-Based Matching: this requires a greater number of vocabular-

ies, which is similar to the entity extraction.
• Machine Learning: if the data seem to be in substantial form, the sta-
tistical analysis can be computed for intensive application. Machine
learning uses the corpus with which it derives the relationships of
the data.
• Hybrid Approach: this is a statistically driven approach but it can
be enhanced through vocabulary. If the content set seems to be a
specific area, then it seems to be the most suitable approach.

10.5.2 Web Mantic Extraction

Web Mantic extraction is the extraction that converts a HTML page into an
XML document. Tables and lists are the two structures which are used to
represent the information of the web pages. It includes few steps, such as:

• A tree generator is used to represent the structures that are stored in

web pages. The lists and tables in the pages create a tree structure.
Patient Report Analysis  145

• For converting HTML to XML, it is the rule generator module where

the data represented in tree form that will be used for generating
rules that represent the important information that has to be trans-
lated. Then, the XML tags are defined by the user.
• The subsumption module is used to generalize the rules which are
defined when an XML tag created will represent the same concept.
• The XML Parser module creates the XML document. Header files of
the XML are automatically written. Semantic generator rules will be
executed sequentially for preprocessing.

10.5.3 Use Cases

• Website Management: in website management the designers tend to

have concern for the management of content, structure, and graphi-
cal presentation simultaneously. As a result, it is difficult to restruc-
ture the websites. This is because integrity constrains enforcement,
and the possibility of creating multiple sites with same data and
updating of the websites. It can be managed by specifying the struc-
ture of the website in a declarative manner as a website exists over
an integrated collection of data. Several systems can be built such as:
Strudel, Araneus, YAT, Autoweb, and Tiramisu.
• Test of XML: XML (Extended Markup Language) is emerging as a
standard for exchanging data over the Web. It enables separation
of the content (XML) and separation of presentation (such as XSL).
Document Type Descriptors (DTDs) provide spatial schemas for
XML documents. All the application then needs is to manage the
XML data. Here, attributes, objects, and atomic values are taken as
tags, elements, and character respectively for extraction of data.

10.6 Unstructured Data
Unstructured data, on the other hand, lacks the organization and precision
of structured data. Examples in this category include physician notes, x-ray
images, and even faxed copies of structured data. In most cases, unstruc-
tured data must be manually analyzed and interpreted.

10.6.1 Finding Meaning in Unstructured Data

It is a great challenge to analyze and interpret unstructured data than struc-
tured data. Free texts and images cannot be categorized easily in the same
way. For example, straightforward code can be used to interpret blood pres-
sure readings (such as normal, elevated, or hypertensive). While considering
146 Machine Learning for Healthcare

the physician’s notes the interpretation of such as “chest pain, gen fatigue,
trouble breath” will also suggest hypertension. However, abbreviations and
spelling errors need to be decoded through human interpretation.
Imagery presents challenges when similar images (such as x-rays and
pathology slides) are indecipherable for all except well-trained professionals
and, even though the clinicians are experienced a second opinion is often
required for validating the diagnosis or interpretation. The unstructured
data is largely analyzed in the medical industry as most of the medical data
requires images to diagnose.
With the advancements of Artificial Intelligence and machine learning
techniques, the unstructured data can be transformed as needed. A natural
language processing tool can be used for decoding the physician’s note, and
the above example then can be interpreted as “chest pain, general fatigue,
trouble breathing”. A machine learning decision support tool might be used
for suggesting that these symptoms are related to hypertension
Data scientists are working with large amounts of data from repositories
for training the machine learning models for pattern recognition of medical
images thereby providing automated secondary opinion for interpretation
or diagnosis of disease.

10.6.2 Extraction of Data
10.6.2.1 Text Extraction

Step 1: it is impossible to analyze the entire text manually, hence a ran-

dom or stratified sample is needed to build a dictionary.
Step 2: for capturing the true essence of the available text, the data
needs to be cleaned. For example, “Sathish’s, Sathish, and Satish”
should be considered as one word. Another important thing is to
remove the stop words.
Step 3: after cleaning the text, the most frequently occurring words are
extracted. Manual identification of the frequently occurring words
can also be conducted. This will form the dictionary.
Step 4: entire datasets has been cleaned which will make sure that the
created dictionary will work on the entire dataset.
Step 5: with this dictionary, every transaction statement can be
categorized.
Step 6: once the tags on each transaction statement are created, the
entirety of the data can be summarized for gaining business strate-
gies and insights.

10.6.2.2 Image Extraction
In the healthcare field, computer vision plays a major role. It is used to ana-
lyze medical-related images and medical research-related PDFs for different
Patient Report Analysis  147

purposes, such as drug development processes, patient care systems, etc.

With computer vision, unusual patterns can be analyzed using AI algo-
rithms for registration, segmentation, and fusion of images.
Data extraction can be taken from two different types of healthcare data
such as:

1. Data from X-rays, CT Scans, MRIs, etc.: the growth rate of such
types of data has been estimated as at 300% each year. These types
of medical data can be used for training computer vision algorithms
for deriving the insights faster to imrpove the patient care. With the
support of AI algorithms, new cures can be realized. It also finds
hidden patterns of diseases and the physicians can thus find more
causes and make more diagnoses of disease.
2. Data from Medical Publications: this includes medical-related the-
ses and information about ongoing medical research and clinical tri-
als. Usually these types of data will be available as PDF. These types
of scanned images cannot be analyzed manually as it is too time-
consuming. Hence, systematic analyzation visualization is necessary.

10.6.2.3 Challenges of Data Extraction from PDFs

Data that are used by the medical researchers will be available on the web
in different formats. It is impossible to extract all the important and relevant
information from PDFs and image files. Though the computer vision uses AI
techniques, it is more challenging to extract data from text images and PDFs
because the document may be in different forms such as single or double
columns, layout variations, multiple fronts, pie charts, and other varying for-
mats (Rosenbloom et al. 2011).
PDFs are scanned images, hence it is hard to read the text and the text
extraction of PDF includes footnotes removal, exclusion of graphs, and so on.
To overcome this problem, optical character recognition is used for differ-
entiating the symbols and formats, etc., for enabling the efficient readability
of the scanned document (Wu et al. 2018). The feature extraction steps are
shown below with an example:

1. Original image of Vessel Extraction from Digital Retinal Images

A sample image is shown in Figure 10.2 for the image analysis discussion.
2. Forming Identity Kernel
The identity kernel will not change the given original image and are
seen as a black and white form as shown in Figure 10.3.
3. Edge Detection (Horizontal Form)
The changes in the horizontal direction of the image can be seen in
Figure 10.4. We can observe that the light that occurs from the light-
ing beam is filtered off.
148 Machine Learning for Healthcare

FIGURE 10.2
Original image of retina.

FIGURE 10.3
Identity kernel formation.

4. Edge Detection (Vertical)

The changes will be made in the vertical direction of the images. It
is observed that the light beam has been returned in blurred form in
the natural image, as shown in Figure 10.5.
5. Edge Detection (Gradient Magnitude)
The square root of summed value is needed in order to get the gradi-
ent magnitude as shown in Figure 10.6.
Patient Report Analysis  149

FIGURE 10.4
Edge detection in horizontal form.

FIGURE 10.5
Edge detection in vertical direction.

6. Edge Detection (Gradient Direction)

Using the arc tangent function, the gradient direction can be
obtained, as shown in Figure 10.7.
7. The Sobel Filter (Gradient Magnitude)
Now let us consider the Sobel filter used for getting the gradient
magnitude as shown in Figure 10.8.
150 Machine Learning for Healthcare

FIGURE 10.6
Edge detection in gradient magnitude.

FIGURE 10.7
Edge detection in gradient direction.

8. Sobel Filters (Gradient Direction)

For gradient directions, the Sobel filter is again used and it can be
found that the roots are clear, as shown in Figure 10.9.
9. Gaussian Blur
Using the Gaussian filter the image is blurred as shown in Figure 10.10
which is called a Gaussian blur.
Patient Report Analysis  151

FIGURE 10.8
Sobel filters in gradient magnitude.

FIGURE 10.9
Sobel filters in gradient direction.

10. Sharpening the Edges

The images can be sharpened by finding the edges of the images as
shown in Figure 10.11.
11. Emboss
With this tool, the images can be viewed with a shadow effect. This
results in forming a “bump map” of the image.
152 Machine Learning for Healthcare

FIGURE 10.10
Gaussian blur.

FIGURE 10.11
Image with sharpened edges.

10.6.2.4 Video Extraction
The video data model is the representation of video data and its content. The
idea behind the model is segmentation or annotation of video. Mining of
video data requires a good data model for the representation. Various mod-
els have been proposed by different authors. Petkovic et al., (2000) proposed
a content-based data retrieval model which includes four layers. They are:
Patient Report Analysis  153

1. A raw video data layer that includes a sequence of frames, as well as

some of the attributes of the video.
2. A feature layer which consists of domain independent features
extracted from the raw data. It also characterizes the colors, shapes,
motion, and textures.
3. An object layer holds the entities which are characterized by prom-
inent spatial dimensions and are then assigned to all the regions
across the frames.
4. An event layer holds the entities which are characterized by a prom-
inent temporal extent which describes the interactions and move-
ments of different objects in a spatial-temporal manner.

10.6.2.5 Sound Extraction
Vacher et al., (2006) state that the everyday sounds that exists around us
can be divided into different classes. The criteria used for this categoriza-
tion includes statistical probability of occurrence of sound in the everyday
life, alarm sounds (maybe a scream) for priority, sound duration (unalarm-
ing sounds are considered to be short and impulsive). These classes can be
related to two different categories:

• Normal sounds that happen which are related to the usual activity
of patients such as doors closing or being locked, walking sounds,
phones ringing, the sound of dishes being moved, and human
sounds such as coughing, and sneezing.
• Abnormal sounds that can be interpreted as a kind of alarm include
breaking glasses, human screams, the sound of someone falling, and
more. If it is recognized, the sound analysis system will transmit an
alarm to the application used for medical supervising. The decision
to call the emergency department is made by this data fusion system.

The signal detection seems to be more important because once an event is lost,
it will be lost forever. On the other hand, with the best conditions the start and
stop time of sounds must be established accurately for using the classifica-
tion steps. Unlike fast fourier transform, wavelet transform seems to be more
adapted for signaling which have more localized features than time indepen-
dent wave like signals such as door slap, breaking of glasses, step sound etc.

10.6.3 Algorithms
10.6.3.1 Natural Language Processing
In the healthcare industry, the clinical information comes in the form of
written text which will be in huge forms, such as laboratory reports, physi-
cal examination reports, operation notes of patients, discharge related
154 Machine Learning for Healthcare

summaries, etc. These are usually in unstructured forms and are not com-
prehensible for the computer-based programs as they need special models
for processing the text (Luo et al. 2016). The Natural Language Processing
model provides a solution to these issues by identifying a series of keywords
that are relevant to the disease in the patient notes based on the existing
databases, thereby enriching the structured data for supporting clinical
decision making.

10.6.3.2 Naïve Bayes
The Naïve Bayes classifier is a probabilistic method used for categorizing the
text, and solving the problem of document predictions for finding the cate-
gory to which it belongs to. The Naïve Bayes classifier considers that one par-
ticular feature of the class will be unrelated to other features. Even though
the features of a class are independent, all its properties will independently
contribute its probability for a certain category. It is one of the most efficient
probabilistic classification algorithms that are successfully applied for many
of the medical related problems.

10.6.3.3 Deep Learning
Deep learning belongs to the machine learning family and it is based on
the artificial neural network techniques, as it is a neural network with an
increased number of layers. When compared to traditional machine learning
algorithms, the more complex non-linear patterns can be learned using the
deep learning algorithms in the data. Modules are pipelined and are train-
able. It is a scalable approach and the automatic feature extraction of data can
be performed.
In healthcare applications, these types of algorithms handle both the tasks
such as machine learning and language processing. The predominantly
used deep learning algorithms are convolution neural networks, deep belief
networks, the multilayer perception model, and recurrent neural networks.
It remains one of the most effective classification algorithms and is success-
fully used in addressing many healthcare-related problems, such as health-
care report classification and journal classification.

10.6.3.4 Convolutional Neural Network

Convolutional neural networks (CNNs) are developed to handle high
dimensional data or data with an increased number of traits. As proposed
by LeCun et al., (1995), the pixel values that are rectified with the normal-
ization of images will be the inputs. Convolutional networks were inspired
by medical processes, and thus the connectivity pattern that exists between
the neurons with separate cortical neurons (which respond to the stimuli in
the region) is restricted. However, the whole visual field is covered as the
Patient Report Analysis  155

receptive field of various neurons will overlap. The CNN then transfers the
weighted pixel values of the image in the convolution layers and sampling is
done in the subsampling layers. The final output will be a recursive function
of the input values.

10.6.3.5 Phenotyping Algorithms
Phenotyping algorithms are implemented using the samples of the diseases
on the EHR data that are usually collected from healthcare units for diagnos-
ing the diseases. The data may be in an unstructured form which contains
large amount of texts from the physicians’ reports, various diagnostics of
diseases, and different vital signs. A phenotyping algorithm is a different
form of special model that is carried through various numbers of medical
data points with specific codes for radiology results, billing, and natural
language processing where different forms of texts are extracted from the
physicians. Machine learning algorithms with supported vector machines
can be applied for identifying arthritis in a combination of patient’s prescrip-
tion records for improving the accuracy of predictive models of disease. For
example, the prevalent condition of diabetic patients can be suggested by
examining the usage of hypoglycemic agents that are collected from the pre-
scription records.

10.6.4 Use Cases

1. Automated Trigger: an automated trigger for sepsis clinical deci-

sion support using ML has been created. It involves the extraction of
text and vital signs for predicting the life-threatening infection that
may affect the patients. Natural Language Processing (NLP) is used
for extracting the data from the clinical text. It has been found that
the area under the curve value is 0.667 without using the NLP and
0.86 when the NLP is used. It was also seen that the accuracy of the
model is increased when the language processing algorithm is used.
2. Patient Risk Prediction: this is an important process as it is used for
making decisions. It assists the physicians in making valuable pre-
dictions. The predicted test result values will be used to ensure that
the particular treatment which was undergone is useful or not. It has
been found that out of the total predictive rules used, 97% of them
seem to be more sensible when the NLP is used. There are other
cases where the physician’s prediction ability is seen to be poor. For
this instance, oncologists found that only 20% accuracy is achieved
while predicting the survival rates of ill patients.
3. Cohort Building: this can be done by leveraging the oncology
department’s electronic health record data. A demonstration of
non-small cell lung cancer is done using the structured as well as
156 Machine Learning for Healthcare

unstructured data. It is found that 8,324 patients are affected with

non-small cell lung cancer by using both the type of data. Out of the
8,324 patients, more than 2,000 patients were found with the cohort
which was formed by structured data. In addition to this, 1,090
patients would be further included in the cohort if only the struc-
tured data is used. It was found that more than 1,000 patients did
not match with the parameters of the study. Hence, only the patients
affected with non-small cell lung cancer that are more than 2,000
in number were found to be the true cohort that can be used for
analysis. This analysis highlights the importance of analyzing both
structured and unstructured data.

10.7 Conclusion
This chapter has outlined that there is a consequential need for the improve-
ment of structured, semi-structured, and unstructured healthcare data for
storing, analyzing, and interpreting. Though powerful tools already exist for
analysis – one that might help the analysts to analyze the data well – there
is a lack of standardization which continues to impede the overall process.
Machine learning, language processing, and Artificial Intelligence have the
potential to streamline the way that the unstructured data can be utilized,
but we fail to capture the point that the machines are making the critical
decisions instead of traditional decision-making physicians. Regardless, all
patients should aexpect and look forward for improved medical or health
outcomes as the technological advancements continue to improve the way
health data are used. Thus, this chapters elaborates on the different forms
of healthcare data with examples of relevant algorithms and use cases,
thereby supporting users to understand the basic concepts of healthcare data
analysis.

References
Asif, Muhammad, H. F. M. C. M. Martiniano, A. M. Vicente, and F. M. Couto
“Identifying disease genes using machine learning and gene functional simi-
larities, assessed through Gene Ontology”. PLoS One 13(12) (2018): 12.
Ba, Mohan, and H. Sarojadevi “Disease diagnosis system by exploring machine
learning algorithms”. International Journal of Innovations in Engineering and
Technology 10(2) (2018): 14–21.
Chen, Min, et al. “Disease prediction by machine learning over big data from health-
care communities”. IEEE Access 5 (2017): 8869–8879.
Patient Report Analysis  157

Istrate, D., E. Castelli, M. Vacher, L. Besacier, and J.-F., Serignat. Information

Extraction From Sound for Medical Telemonitoring. Information Technology
in Biomedicine, IEEE Transactions on, 2006, 10(2), pp. 264–274. ffhal-00419915f
Jutel, Annemarie “Classification, disease, and diagnosis”. Perspectives in Biology and
Medicine 54(2) (2011): 189–205.
Kaur, Harleen, and Siri Krishan Wasan “Empirical study on applications of data min-
ing techniques in healthcare”. Journal of Computer Science 2(2) (2006): 194–200.
Y, LeCun and Y. Bengio, Convolutional networks for images, speech, and time series,
The handbook of brain theory and neural networks 3361 (10) 1995.
Luo, Jake, M. Wu, D. Gopukumar, and Y. Zhao “Big data application in biomedical
research and health care: A literature review”. Biomedical Informatics Insights 8
(2016): BII-S31559.
Petkovic, M., and Jonker, W. An Overview of Data Models and Query Languages for
Content-based Video Retrieval. (2000).
Razia, Shaik, et al. “A review on disease diagnosis using machine learning tech-
niques”. International Journal of Pure and Applied Mathematics, 117(16), (2017):
79–85.
Rosenbloom, S. Trent, et al. “Data from clinical notes: A perspective on the tension
between structure and flexible documentation”. Journal of the American Medical
Informatics Association JAMIA 18(2) (2011): 181–186.
Sharmila, S., C. Leoni, Dharuman, and P. Venkatesan “Disease classification using
machine learning algorithms-A comparative study”. International Journal of
Pure and Applied Mathematics 114(6) (2017): 1–10.
Vinitha, S., et al. “Disease prediction using machine learning over big data”. Computer
Science & Engineering: Anais an International Journal (CSEIJ) 8 (2018): 1.
Wu, Honghan, et al. “SemEHR: A general-purpose semantic search system to surface
semantic data from clinical notes for tailored care, trial recruitment, and clini-
cal research”. Journal of the American Medical Informatics Association JAMIA 25(5)
(2018): 530–537.
11
Statistical Analysis of the Pre- and
Post-Surgery in the Healthcare Sector
Using High Dimension Segmentation

Soobia Saeed, Afnizanfaizal Abdullah, N. Z. Jhanjhi,

Memood Naqvi, and Mamoona Humayun

CONTENTS
11.1 Introduction................................................................................................. 159
11.2 Methodology .............................................................................................. 161
11.2.1 Sampling Techniques..................................................................... 162
11.2.2 Sample Data and Size..................................................................... 162
11.2.3 Light Field Toolbox for MATLAB................................................. 162
11.2.4 High Dimensional Light Field Segmentation Method.............. 163
11.3 Support Vector Machine (SVM)................................................................ 163
11.3.1 4-Dimentional SVM Graphs.......................................................... 166
11.4 Statistical Technique................................................................................... 166
11.5 Result and Discussion................................................................................ 168
11.6 Conclusion................................................................................................... 170
11.7 Future Work................................................................................................. 171
References.............................................................................................................. 171

11.1 Introduction
Cerebrospinal fluid (CSF) is the fluid that travels through the brain’s ven-
tricles (cavities or voids) and around the surface of the brain and spine. CSF
is one of the most challenging neuro-chirurgical complications (Saunders
et al. 2018).
CSF leakage is a condition that occurs when the CSF leaks through a defor-
mity in the dura or head and exits through the nose or ear. CSF leakage is the
result of a hole or tears of the dura and is the most extreme layer of menin-
gitis. The causes behind the hole or tear can damage the head and the func-
tionality of the brain or breast. CSF slots can occur in the same way in the
lower back, in what is also called spinal cord or spinal anesthesia. A CSF leak

159
160 Machine Learning for Healthcare

can occur without restriction in a similar manner without a known cause

(Suman et al. 2017). The cells that make up these interfaces are also sites of
integrated exchange mechanisms (vectors) that control the entry and exit of
the brain into a wide range of molecules (Shree, Varuna, and Kumar, 2018).
An important mechanism for controlling the distinctive synthesis of inter-
stitial fluid in the brain is the secretion of cerebrospinal fluid by the placental
plexus, which flows through the ventricular system and exchanges materi-
als between the cerebrospinal fluid and the brain (Shree and Kumar 2018).
It is necessary to understand how complex the barrier mechanisms are to
assess the effects of inflammatory conditions in the brain, both in adults and
during development. The exhaustion of the cerebrospinal fluid may occur
through a leakage, a shunt, inadequate production, or very rapid absorp-
tion (Abdullah et al. 2011). There are also some similar syndromes in which
intracranial compliance is very high; they cause similar symptoms when the
brain shrinks when standing up and floats. Cerebrospinal fluid is widely
targeted to detect molecules for cancer detection. This research examines
current scientific knowledge about the biochemical factors in CSF that have
been reported in brain cancer literature (Saeed, Abdullah, and Jhanjhi 2020).
A brain tumor is a mass of unwanted abnormal cells growing in the brain.
Your skull is very rigid in order to protect your brain. Any growth can cause
problems in this narrow space (Lavanyadevi et al. 2017). Cancerous (malig-
nant) or non-cancerous (benign) cells can be found in brain tumors. They
may increase intracranial pressure when tumors grow, both benign and
malignant (Abdullah, Afnizanfaizal, et al, 2011). This can cause damage to
the brain and endanger life. Tumors of the brain are classified as either pri-
mary or secondary (Li et al. 2015).
Many primary tumors of the brain are beneficial. Secondary brain tumors,
also known as diffuse brain tumors, occur when other organs, such as the
lung or breast, spread cancer cells to the brain (Lee et al. 2005). Brain and
spinal cord is uncontrollable groups of abnormal cells in the brain or spinal
cord (Saeed, Afnizanfaizal and Jhanjhi 2019a). It is very important to distin-
guish between benign (non-cancerous) tumors and malignant (cancerous)
tumors in most other parts of the body. Tumors do not grow in nearby tis-
sues or spread to remote areas, so they do not endanger life in other parts of
the body (Saeed and Abdullah 2020). One of the major causes of malignant
tumor severity is that it can spread all over the body. Section one provides
the overview on detection of brain cancer and CSF as an image segmenta-
tion process and an abstract definition of all kinds of segmentation (Saeed
and Abdullah 2019). This research closes with a discussion of the challenges
of the deep learning method in the context of medical imaging and research
issue (Khotanlou et al. 2009). The researcher describes the background of
brain cancer using Interfacing of MRI-4-dimensional image segmentation
(Liang et al. 2018). The researcher describes the details of brain cancer and
explains cerebrospinal fluid (CSF) leakage with the aim of treating cancer
and show the experimental result after the implementation of segmentation
High Dimension Segmentation 161

process and K-NN algorithm as well. Section two presents the methodology
used in the detection of brain cancer using Interfacing of MRI-4D Images.
The sampling techniques and tools will generate experimental result after
the implementation of the statistical analysis. Section three explains the
Support Vector Machine (SVM) results and SVM linear equation. Section
four describes the method used in Statistical Technique. The framework
can identify cancer damage areas or isolated tumors and non-tumors well
by using 4D image light field segmentations. Initially, the MRI processes
the pre-processed image method with the final target selected to adjust the
image for the rest of the procedures. Section five provides the details of
results and discussion with comparison of previous result to new achieve-
ments. Section six presents the conclusion, describes the contributions made
by this study, and suggests future directions. This chapter also presents the
achievements of the set objectives and the comparative performance evalu-
ations and simulation of experimental results. One of the contributions of
this research is presenting the results of statistical data. The researcher ana-
lyzes the all results in 2D and 3D formats, but we didn’t find any available
work on 4D platform to analyze the statistical data of patients. In addition,
this research differs from previous work by presenting MRI tool analysis
in detail.

11.2 Methodology
The proposed research aim is to acquire statistics related to the detection of
most brain cancers (because of growing via CSF leakage) with the interface
of MRI high dimensional images. The researcher discusses the damaged
cells of the mind due to the abnormalities of the mobile. This is a qualitative
research study. Therefore, it is a part of a large number and uses secondary
sources, which includes comparative research studies from a global perspec-
tive. The primary aim of our work is to build up a framework that under-
stands the tumor area. At first, the data MRI picture is preprocessed with a
specific stop aim to restore the photo for relaxation of the procedures with
the help of light subject toolbox of excessive dimensional and their graphical
illustration.
The sampling collections are used for the have a look at consists of clinical
mind snap shots for human being. The researcher will almost implement and
simulate the experiments of (4D picture segmentation procedure) to the test-
ing on human being and their mind tissues (animal mind pattern) randomly
from Malaysia. The major goal of our proposed model is to broaden a gadget
that could come across the CSF leakage inside the brain and tumors place
or can separate between tumors and non-tumors patient. Data are accrued
thru various assets, which are applying the necessary research equipment.
162 Machine Learning for Healthcare

SVM tool is one of the important equipment to for you to teach the facts
and tested the original sampling records. The SVM technique is one of the
not unusual approaches to satisfy this requirement and boom the margin
among classes, so that the overall overall performance is generally higher.
Regarding the usage of more than one classes in SVM, comparison assess-
ment is used. Thus, the researcher uses SVM to decide the objectivity of the
classifications. This version identifies the most cancers at very early degree.
Moreover, the clinical nice of relevant literature could be checked through
important appraisal tools for higher evaluation.

11.2.1 Sampling Techniques
Sampling techniques are usually collected from the 4D light field segmen-
tation process. The purpose of this study is to detect brain cancer through
MRIs of the brain, for example, using supervised machine learning in rela-
tion to 4D light field segmentation, and SVM (support vector machine) tools
for training and testing of datasets. The SVM approach is one of the most
common ways to meet this requirement. SVMs increase the margin between
categories so that the overall performance is generally higher. Regarding the
use of multiple categories in SVM, comparability is used and SVM is used to
determine the objectivity of the classifications.

11.2.2 Sample Data and Size

The sampling data is based on human tissue and scans images from MRIs.
The sample size population is more than 200 (for the damaged brain cell
after the CSF leakage and selecting the sample for human brain images or
CSF sample images as well). The data was entered and analyzed using SPSS
version 19.0. The main areas found to be ones of concern were the impor-
tance of using brain language and time management, and identifying the
main area of patient concerns.

11.2.3 Light Field Toolbox for MATLAB

The purpose of using a set of tools to work with light field images (also
known as plenoptic images) in MATLAB is to decode, calibrate correct, cor-
rect color, filter basic images, and visualize light field images. There are sev-
eral ways to represent the 4D lighting field segmentation. The researcher
adopted the Lumigraph method to clarify the rays in the three-dimensional
space. The intersection points with “u-v” and “x-y” are defined in the 3D
coordinates of the beam (rays). The radius can be represented as a point at a
4D distance such as p = (u, v, x, y), and the intensity of p is represented. The
representation lumigraph can be converted to multiple representations that
contain a viewpoint level, an image plane, and vice versa. In multi-point dis-
play, the U-V and X-Y models, respectively, correspond to the artifact s width
High Dimension Segmentation 163

and image. In this thesis, the researcher explains the multifactorial method
in a way that is easy to understand.

11.2.4 High Dimensional Light Field Segmentation Method

The purpose of selection of a light field is basically a 4D structure. Most of
the sounds of each pixel corresponding to the ray, the two dimensions deter-
mine the position of that ray, while the other two determine its direction. In
the case of images measured by a camera, based on a camera lens such as
Lytro, the two dimensions choose a lens image. The remaining two of them
choose pixels within the image of this lens to maintain the resolution of the
segmentation.
Table 11.1 shows that the original images of 3D images of the CSF leakage
in the spinal cord. This is converted in to the 4D image images and create the
histogram for finding the accuracy and size of CSF leakage in the spinal cord
and, additionally, the graphical representation has been programmed as you
can see in the second image of the CSF.
Leakage here is more prominent in the area of the tumor when compared
to the previous image. After implementation of the 4D image process, the
image is converted into grayscale and it clearly displays the size of the CSF
tumor – which is 0 – so it means that the statistical graph shows the size of
the tumor is 2,550. Here another value of maximum intensity (max_c) repre-
sents the intensity of the CSF-tumor which is 15,023 and shows the appear-
ance of CSF leakage. Image 2 show the size of the CSF-tumor which is still
in 0 but the value of statistical graph is different at 2,550. This shows that the
situation of CSF leakage – which is slowly increasing in the brain and dam-
aging the brain cell – and the max_c of Image 2 is 887. Image 3 also shows
that the size of the CSF-tumor is 0 and the statistical graph shows the size of
tumor value is 2,550 but that the max_c is 2,621 which represent the tumor
shell in hard form after the CSF has been deposited in the brain. This is basi-
cally the leakage from initial to final stage of the tumor in the spinal cord
which can be seen in Table 11.1.

11.3 Support Vector Machine (SVM)

The SVM monitors a learning algorithm based on the theory of statistical
learning. In view of the sample data, the named dataset (training set), D =
{Q, p,x class}, SVM tries to calculate the function ‘y’ f designation as f(x) = y
for all samples in the dataset. The assignment function describes the rela-
tion between data samples and their class labels; new, unknown data are
classified. In the context of the SVM, classification is carried out using the
following classification function (a process called the advanced stage). The
TABLE.11.1
164

4-Dimention Size of CSF Leakage in Spine

IMAGE 1 IMAGE 2 IMAGE 3
Machine Learning for Healthcare

(Continued)
TABLE.11.1 (CONTINUED)
4-Dimention Size of CSF Leakage in Spine
IMAGE 1 IMAGE 2 IMAGE 3
High Dimension Segmentation

Size of CSF leakage in Spine Size of CSF leakage in Spine Size of CSF leakage in Spine
max_c =15023 max_c =887 max_c =2621
165
166 Machine Learning for Healthcare

FIGURE 11.1
Statistical graph of SVM.

purpose of support vector machine is to explore the idea of transforming an

input field into a high-dimensional space in order to optimize the best clas-
sification function, otherwise, and SVM is able to run RBF and view multiple
layers. Figure 11.1 presents SVM graph.

11.3.1 4-Dimentional SVM Graphs

Here we show the experimental results of SVM classification after the train-
ing and testing of the given dataset.

y = X*x+Y (11.1)

Here X represents the value of trained datasets and Y represents the value of
testing datasets after the implementation of SVM classification techniques.
The values of X and Y are given below:
Coefficients variable values of data sets of SVM:

X = 0.29308

Y = 1.062

Norm of residuals = 2.0304

11.4 Statistical Technique
Google gathered the data source, which is developed, structured, and verified
by the American Cancer Society and multiple cancer and CSF surgery-related
High Dimension Segmentation 167

hospitals. The data was collected across the period of 2018–19 and, after-
wards, responses were feed into the SPSS “Statistical Package for compu-
tational research” for further analysis. In SPSS the researcher used a t-test
on collected data for evaluation. In this study, data was analysed through
SPSS. SPSS is a statistical tool which is commonly used by many research-
ers. The t-test was applied to find the relationship between MRI sequences
for brain cancer with CSF leakage. The t-test is used for finding the signif-
icant ratio, probability, and accuracy of MRI sequence results. The results
of pre- and post-surgery are given more priority than normal test results.
T-tests estimate the results involving all the above-mentioned independent
variables with the dependent variable. This model shows strong relationship
between dependent and independent variables based on the probability of
findings; the outputs are interpreted. Since the result is based on a two-tailed
approach, this indicates that there will be difference between the means, but
the direction of the difference can’t be predicted (Table 11.2).
Figure 11.2 shows the statistical results in terms of graphical represen-
tation as the authors list the complete details of pre- and post-surgery of
patient results, and their tool usage. As we can see that the values of all tools
are mentioned, as is the accuracy of MRI significant tools. The same situa-
tion varies for Table 11.3 after the implementation of SPPS test results and

TABLE 11.2
Statistical Results of Pre- and Post-Surgery Data
Group Statistics
Std. Error
Surgery N Mean Std. Deviation Mean
Pre-Surgery 22 .73 .456 .097
Post-Surgery 22 .00 .000 .000
RAVLT Pre-Surgery 22 .36 .492 .105
Post-Surgery 22 .14 .351 .075
TrailA Pre-Surgery 22 .91 .294 .063
Post-Surgery 22 .55 .510 .109
TrailB Pre-Surgery 22 .91 .294 .063
Post-Surgery 22 .55 .510 .109
HPTR9 Pre-Surgery 22 .86 .351 .075
Post-Surgery 22 .50 .512 .109
HPTL9 Pre-Surgery 22 .91 .294 .063
Post-Surgery 22 .55 .510 .109
COWAT Pre-Surgery 22 .45 .510 .109
Post-Surgery 22 .23 .429 .091
WDRT Pre-Surgery 22 .91 .294 .063
Post-Surgery 22 .55 .510 .109
mwalk10 Pre-Surgery 22 .86 .351 .075
Post-Surgery 22 .55 .510 .109
168 Machine Learning for Healthcare

FIGURE 11.2
Graphical results of pre- and post-surgery data.

also when finding the significant values through using a t-test. Figure 11.3
shows the graphical results of independent samples of pre- and post-surgery
patient data and also shows the range of the majority samples result.

11.5 Result and Discussion

The aim of this research is to construct a proposed framework that can iden-
tify cancer damaged areas or be isolated from tumors and non-tumors and
leakage of CSF. By using HD image light field segmentation with two dif-
ferent forms, the MRI processed the pre-processed image method with the
final target selected to adjust the image for the rest of the procedures. The
basis of this study is the detection of brain cancer and CSF leakage through
to the process of interfacing the 4D image segmentation process. Therefore,
it consists of primary and secondary sources after a research study, followed
by MATLAB software modelling techniques with the use of original medi-
cal sample images to measure the range of brain damage cells deep inside
of CSF.
In this study, the researcher proposes a HD modulation method that super-
vises the light field that can be used to emit light. Depending on the user’s
input and using SVM to maintain the HD frequency (redundancy) light
fields. The researcher uses the technique for editing the brain skull damaged
TABLE 11.3
Statistical Results of Independent Sample Test
Independent Samples Test
Levene’s Test for
Equality of Variances T-test for Equality of Means
95% Confidence
Interval of the
Difference
Sig. Mean Std. Error
F Sig. T Df (2-tailed) Difference Difference Lower Upper
High Dimension Segmentation

NART Equal variances assumes 80.640 .000 7.483 42 .000 .727 .097 .531 .923
Equal variances do not assume 7.483 21.000 .000 .727 .097 .525 .929
RAVLT Equal variances assumes 13.644 .001 1.763 42 .085 .227 .129 −.033 .487
Equal variances not assumes 1.763 37.978 .086 .227 .129 −.034 .488
Trail A Equal variances assumes 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Equal variances not assumes 2.898 33.600 .007 .364 .125 .109 .619
Trail B Equal variances assumes 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Equal variances not assumes 2.898 33.600 .007 .364 .125 .109 .619
HPTR9 Equal variances assumes 23.579 .000 2.748 42 .009 .364 .132 .097 .631
Equal variances not assumes 2.748 37.192 .009 .364 .132 .096 .632
HPTL9 Equal variances assumes 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Equal variances not assumes 2.898 33.600 .007 .364 .125 .109 .619
COWAT Equal variances assumes 8.090 .007 1.600 42 .117 .227 .142 −.059 .514
Equal variances not assumes 1.600 40.810 .117 .227 .142 −.060 .514
WDRT Equal variances assumes 40.000 .000 2.898 42 .006 .364 .125 .110 .617
Equal variances not assumes 2.898 33.600 .007 .364 .125 .109 .619
mwalk10 Equal variances assumes 22.120 .000 2.411 42 .020 .318 .132 .052 .584
Equal variances not assumes 2.411 37.277 .021 .318 .132 .051 .585
169
170 Machine Learning for Healthcare

FIGURE 11.3
Graphical results of independent samples

images of brain cancer samples. These findings show the effectiveness of our
approach to light editing applications. These light field methods can be use-
ful for improving the quality of the segmentation of application editing and
the composite light field pipeline, as they reduce boundary artefacts. The
statistical data show the accuracy of pre and post-surgery results of brain
cancer with CSF leakage.

11.6 Conclusion
The research closes with a discussion of the challenges of the deep learn-
ing method in relation to medical imaging and the research problem. The
researcher describes the statistical results of brain cancer through an inter-
face of MRI-four-dimensional photo segmentation. The researcher mentions
the details of brain cancer and explains the cerebrospinal fluid (CSF) leak-
age with the concept of treating cancer so as to assist the experimental end
result after the implementation of segmentation process and Tt-checking out
of massive consequences as well.
The researcher also mentions the pre- and post-surgical results with the
usage of SPSS tests and shows the substantial testing of the results. Some of
the tumor’s characteristics are detected and will be useful in medical appli-
cations. The findings of statistical data are one of the most important aspects
of our study. The researcher analyzes all findings in the both 2D and 3D con-
text. However, we have not found any work in the literature on the 4D system
to be used to analyze patient data statistics. Furthermore, the researcher cre-
ates the novelty of MRI images tools and compare the previous work on MRI
tools is available.
High Dimension Segmentation 171

11.7 Future Work
Future work is suggested to include segmenting and detecting more images
with more features that help classify multiple tumor types.

References
Abdullah, Afnizanfaizal, et al. “An improved local best searching in particle swarm
optimization using differential evolution”. In: 2011 11th International Conference
on Hybrid Intelligent Systems (HIS). IEEE, Malaysia2011.
Abdullah, Afnizanfaizal, et al. “Cerebrospinal fluid pulsatile segmentation-a
review”. In: The 5th 2012 Biomedical Engineering International Conference. IEEE,
2012.
Abdullah, Afnizanfaizal, et al. “An improved swarm optimization for parameter
estimation and biological model selection”. PLoS One 8(4) (2013): 4.
Das, Suman, et al. “Detection and area calculation of brain tumour from MRI images
using MATLAB”. International Journal 4 (2017): 1.
Gamage, P. T., and Dr Lochandaka Ranathunga “Identification of brain tumor using
image processing techniques”. Faculty of Information Technology, University of
Moratuwa (2017). https://www. researchgate. net/publication/276133543.
Gelb, Sivan, et al. “Mechanisms of neuropsychiatric lupus: The relative roles of
the blood-cerebrospinal fluid barrier versus blood-brain barrier”. Journal of
Autoimmunity 91 (2018): 34–44.
Gupta, Anjali, and Gunjan Pahuja. “Hybrid clustering and boundary value refine-
ment for tumor segmentation using brain MRI”. In: IOP Conference Series:
Materials Science and Engineering, Vol. 225, No. 1. IOP Publishing, 2017.
Heinen, Rutger, et al. “Robustness of automated methods for brain volume measure-
ments across different MRI field strengths”. PLoS One 11(10) (2016): 10.
Jian, Wen-xuan, et al. “Potential roles of brain barrier dysfunctions in the early stage
of Alzheimer’s disease”. Brain Research Bulletin 142 (2018): 360–367.
Kant, Shawn, et al. “Choroid plexus genes for CSF production and brain homeostasis
are altered in Alzheimer’s disease”. Fluids and Barriers of the CNS 15(1) (2018): 34.
Khan, A. Raouf, Noor Zaman, and Saira Muzafar “Health hazards linked to using
mobile cellular phones”. Journal of Information and Communication Technology 2(2)
(2008): 101–108.
Khotanlou, Hassan, et al. “3D brain tumor segmentation in MRI using fuzzy clas-
sification, symmetry analysis and spatially constrained deformable models”.
Fuzzy Sets and Systems 160(10) (2009): 1457–1473.
Kinaci, Ahmet, et al. “Effectiveness of dural sealants in prevention of cerebrospinal
fluid leakage after craniotomy: A systematic review”. World Neurosurgery 118
(2018): 368–376.
Lavanyadevi, R., et al. “Brain tumor classification and segmentation in MRI images
using PNN”. In: 2017 IEEE International Conference on Electrical, Instrumentation
and Communication Engineering (ICEICE). IEEE, 2017.
172 Machine Learning for Healthcare

Lee, Chi-Hoon, et al. “Segmenting brain tumors with conditional random fields and
support vector machines”. In: International Workshop on Computer Vision for
Biomedical Image Applications. Springer, Berlin, Heidelberg, 2005.
Lee, Su Yeon, et al. “Regulation of tumor progression by programmed necrosis”.
Oxidative Medicine and Cellular Longevity, 2018, 1–28,(2018).
Li, Guodong, et al. “Automatic liver segmentation based on shape constraints and
deformable graph cut in CT images”. IEEE Transactions on Image Processing : A
Publication of the IEEE Signal Processing Society 24(12) (2015): 5315–5329.
Liang, Fan, et al. “Abdominal, multi-organ, auto-contouring method for online adap-
tive magnetic resonance guided radiotherapy: An intelligent, multi-level fusion
approach”. Artificial Intelligence in Medicine, 90 (2018): 34–41.
Liu, Jia, et al. “A cascaded deep convolutional neural network for joint segmenta-
tion and genotype prediction of brainstem gliomas”. IEEE Transactions on Bio-
Medical Engineering 65(9) (2018): 1943–1952.
Mendrik, Adriënne M., et al. “MR BrainS challenge: Online evaluation framework
for brain image segmentation in 3T MRI scans”. Computational Intelligence and
Neuroscience 2015, Nagoya, Japan, 18.(2015).
Miller, Kimberly. Cancer Treatment & Survivorship Facts & Figures. American Cancer
Society, pp.1–44, 2016.
Moeskops, Pim, et al. “Automatic segmentation of MR brain images with a convo-
lutional neural network”. IEEE Transactions on Medical Imaging 35(5) (2016):
1252–1261.
Oliveira, Gustavo Casagrande, Renato Varoto, and Alberto Cliquet Jr. “Brain tumor
segmentation in magnetic resonance images using genetic algorithm cluster-
ing and adaboost classifier”. BIOIMAGING.2, 77–82. (2018).
Prahl, Louis S., et al. “Glioma cell migration in confined microchannels via a motor-
clutch mechanism”. bioRxi 1(2), 1 -12, (2018): 500843.
Saeed, Soobia, and Afnizanfaizal Bin Abdullah “Investigation of a brain cancer with
interfacing of 3-dimensional image processing”. In: 2019 International Conference
on Information Science and Communication Technology (ICISCT). IEEE, 2019.
Saeed, Soobia, and Afnizanfaizal Abdullah “Recognition of brain cancer and cere-
brospinal fluid due to the usage of different MRI image by utilizing support
vector machine”. Bulletin of Electrical Engineering and Informatics 9(2) (2020):
619–625.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi “Analysis of the Lung
Cancer patient’s for Data Mining Tool.” IJCSNS 19(7) (2019a): 90.
Saeed, Soobia, Afnizanfaizal Abdullah, and N. Z. Jhanjhi “Implementation of Fourier
transformation with brain cancer and CSF images”. Indian Journal of Science and
Technology 12(37) (2019b): 37.
Saeed, Soobia, and Raza Jafri “Estimation of brain tumor using latest technology of
mobile phone”. Journal of Information and Communication Technology (JICT) 9(1)
(2015): 32–09.
Saeed, Soobia, and Shahbaz Ahmed Noor “Analysis of a brain tumour due to the
usage of mobile phone”. Mehran University Research Journal of Engineering &
Technology Journal 36(3): 609–620, July 2017.
Saunders, Norman R., et al. “Physiology and molecular biology of barrier mecha-
nisms in the fetal and neonatal brain”. The Journal of Physiology 596(23) (2018):
5723–5756.
High Dimension Segmentation 173

Shree, N. Varuna, and T. N. R. Kumar “Identification and classification of brain

tumor MRI images with feature extraction using DWT and probabilistic neural
network”. Brain Informatics 5(1) (2018): 23–30.
Usman, Khalid, and Kashif Rajpoot “Brain tumor classification from multi-modality
MRI using wavelets and machine learning”. Pattern Analysis and Applications
20(3) (2017): 871–881.
van der Kleij, Lisa A., et al. “Fast CSF MRI for brain segmentation; Cross-validation
by comparison with 3D T1-based brain segmentation methods”. PLoS One 13(4)
(2018): 4.
12
Machine Learning in Diagnosis
of Children with Disorders

Lokesh Kumar Saxena and Manishikha Saxena

CONTENTS
12.1 Introduction................................................................................................. 175
12.1.1 Down Syndrome (DS).................................................................... 176
12.1.2 Sensory Processing Disorder (SPD)............................................. 176
12.1.3 Autism Spectrum Disorder (ASD)................................................ 176
12.1.4 Aims and Organisation................................................................. 177
12.2 Existing Tools for Diagnosis of DS, SPD, and ASD................................ 178
12.2.1 Existing Tools of DS Diagnosis..................................................... 178
12.2.2 Existing Tools of SPD Diagnosis................................................... 179
12.2.3 Existing Tools for ASD Diagnosis................................................ 179
12.3 Machine Learning Applied for Diagnosis of DS, SPD, and ASD........ 180
12.4 Machine Learning Case Studies of DS, SPD, and ASD......................... 180
12.4.1 Machine Learning (ML) Case Study for DS................................ 180
12.4.2 Machine Learning Case Study of SPD......................................... 182
12.4.3 Machine Learning Case Study for ASD...................................... 182
12.5 Conclusion................................................................................................... 183
References.............................................................................................................. 183

12.1 Introduction
Sustainable development refers to a development fulfilling the present needs
of the world and preserving the resources for future generations (Bruntland
1987). It has three pillars: economic, social, and environment (Elkinton
2004). The European Commission has taken it on as a fundamental objec-
tive (Commission 2009). Now, industries are pressured to include sustainable
development in their business goals, but the social dimension has been left
unaccounted for by industry (Seuring 2013; Lokesh et al. 2016, 2018a,b, 2020).
The European Commission declared in its agenda for 2014–2020 that its prime
goal is to foster economic growth with regional development, coupled with job
creation. To participate in economic growth, the people must have appropriate

175
176 Machine Learning for Healthcare

education and training, as well as adequate health. A particular section of soci-

ety, are known as disabled children (or people who did not get opportunities
due to a lack of appropriate education, training, and health and therefore did
not contribute in the economic growth of nation, society, or themselves). These
children had a disorder that developed either during pregnancy (prenatal) or
after birth. These disorders prevent them getting appropriate education, train-
ing, and wellbeing. In recent years, the number of children with disabilities
has increased at an exponential rate. Common disabilities include neurodevel-
opmental disorders such as Down’s syndrome (DS), Autism spectrum disor-
der (ASD), and sensory processing disorder (SPD).

12.1.1 Down Syndrome (DS)

More than 1 in 1,000 newborn babies have DS, a disability requiring con-
siderable social, monetary, and legal help. Almost 85% infants with Down’s
syndrome live past their first year, and about 50% of survivors have a life
expectancy of over 50 years. DS is a major chromosomal disability in fetuses.
It has an 8% share in all congenitally abnormal newborn babies (Koivua
et al. 2018).

12.1.2 Sensory Processing Disorder (SPD)

SPD is a clinical state which makes it difficult to modulate and organize sen-
sory inputs and employ them adequately. About 16% children in the United
States suffer with SPD (Parabases et al. 2019). Children who have SPD usu-
ally lack the skills needed for educational achievement in schools. Owing to
the sensory processing disruption, a lot of children who have SPD show late
language and motor development progress. These are shown clearly over
time with higher complexity. Thus, children who have SPD usually face edu-
cational, emotional, and social obstacles. It includes poor self-concept, anxi-
ety, inattention, aggression, dysgraphia, and educational failures.
Furthermore, there are some children with SPD who also have common
sensory problems such as hyper- or hypo-sensitivity to sound and touch.
These children might show other behavioral conditions like ASD and atten-
tion deficit hyperactivity disorder (ADHD). Most children who have SPD do
not display social communication problems as often found in ASD. But, up
to 92% of persons with ASD display hyper- or hypo-sensitivity. ADHD indi-
caors are usually found in more than 40% children who have SPD, such as
acute inattention and/or hyperactivity (Payabvash et al. 2019a).

12.1.3 Autism Spectrum Disorder (ASD)

ASD is a neuron development-related disorder which limits communica-
tion, social transactions, and behavior. It is a rapidly growing in the world.
Its causes are associated with genetics and neurologic factors. It is mainly
Diagnosis of Disordered Children 177

detected by utilizing non-genetical factors such as communication, repeti-

tive behaviors, social transactions, play, and imagination. Almost 1.5% the
world population is thought to be on the Autistic spectrum. Recent studies
have reported a 3:1 ratio of male-to-females with ASD. In the US, about 1 in
59 children, are reported to have ASD.
ASD prevalence in US children has grown by 120% since the year 2000.
ASD has been listed among the fastest-growing developmental disabilities.
It costs about $250 billion per year in for US healthcare (Center for Disease
Control and Prevention 2014). ASD may vary in severity ranging from mild,
to somewhat limiting of normal life, to a severe disability requiring insti-
tutional care. Unfortunately, at present, no one single diagnostic test is able
to identify ASD accurately. Instead, diagnosis is usually made by special
experts through a comparative evaluation of a person’s abnormal behaviors
when compared to other children of the same age (Lord and Jones 1994, 2000).
ASD is a lifelong disability. Thus, early identification to treat is very impor-
tant in order to enhance the life of child and their family, and to decrease the
financial burden on society.
ASD affects the day-to-day life of the individual. It creates considerable
stress for their families. The traits are usually displayed when the child is
three. Environment and genetics-related factors are supposed as a major
causal factor for ASD. An important autistic characteristic is difficulties in
social interactions. Here, the autonomic nervous system plays a key role.
Other autistic traits include having limited interests and rigid or repetitive
behaviors. The main communication gap can sometimes be seen through no
or unsteady eye contact, slow or no response to somebody speaking, placing
all focus on a favorite subject without caring for others’ interests, an unusual
voice tone, and the inability to understand other people’s activities. People
with ASD can also exhibit echolalia or repetitive behaviors. They can exhibit
keen interest in some topics, sometimes facts or numbers. They tend to be
upset by tiny changes in daily routine. They may have less or more sensitiv-
ity than other people to sensory inputs such as light or noise, etc. Autistic
people may also possess many strengths, which sometimes include such
things as extremely improved memory, learning ability for the finer details
of things, advanced visual or auditory comprehension, or to excel in sciences,
or mathematics, or music.

12.1.4 Aims and Organisation

This chapter is aims to provide the following insights:

1. To provide a short overview of research with the hope of motivat-

ing engineering and medical professionals to conduct collaborative
research and develop technological solutions for diagnosis and the
medical care of children/persons with disabilities using machine
learning.
178 Machine Learning for Healthcare

2. To identify the machine learning tools needed for such collaborative

research.
3. To report on existing research with novel application of machine
learning to the prenatal and postnatal early diagnosis of some dis-
abilities such as Down’s syndrome, Autism disorder, and Sensory
Processing Disorder to prevent occurrence and/or to provide early
medical care to facilitate children in getting the education and train-
ing needed to become productive contributors in the growth of
themselves, the nation, and world.

This chapter is organized into five sections. Section 1 comprises of a pream-

ble and introduction. Section 2 shows existing tools for diagnosis, and prob-
lems facing DS, SPD, and ASD. Section 3 describes machine learning applied
for diagnosis of DS, SPD, and ASD. Section 4 describes machine learning
case studies of DS, SPD, and ASD. Section 5 offers a conclusion.

12.2 Existing Tools for Diagnosis of DS, SPD, and ASD

This section discusses existing diagnostic tools for DS, SPD, and ASD.

12.2.1 Existing Tools of DS Diagnosis

Risk screening of first trimester for Down’s syndrome (also known as tri-
somy 21, or T2) is usually done by examining maternal attributes and many
biomarkers, e.g. FhCGβ (maternal serum free human chorionic gonadotropin
β-subunit); PAPP-A (pregnancy associated plasma protein-A); and NT (ultra-
sound nuchal translucency). A first trimester test with above three markers
is known as a combined test (Cuckle and Maymon 2016).
The existing pre-natal screening software for DS risk evaluation – e.g.
LifeCycle™ from PerkinElmer, Waltman, MA, USA – uses the multi-vari-
able based Gaussian models. Risk evaluation methods use a risk evaluation
engine to compute the risk amounts. Basic building elements in protocol are
some functions in the protocol. The marker results obtained from the model
of risk are first normalized, and then the required rectifications are included
for MoM (modified multiple of median) values of every marker. Then, the
probability of T21 is computed as a ratio an affected likelihood to unaffected
likelihood. The likelihood for effected and unaffected situations are found
using a multi-variable GDF (Gaussian density function). GDF employ the
population measures for each marker i.e. mean, SD (standard deviation), and
marker correlation for given gestation and computed MoM value of every
marker. For LifeCycle™ version 4.0, the population measures were taken
from the study by Wald et al. (2003).
Diagnosis of Disordered Children 179

12.2.2 Existing Tools of SPD Diagnosis

The diffusion tensor imaging (DTI) micro structural is used for SPD diagno-
sis. In DTI, correlates/matrices of SPD for identification of children are MD
(mean diffusivity), FA (fractional anisotropy), AD (axial diffusivity), and RD
(radial diffusivity).
DTI research in SPD children found defective microstructures of white
matter primarily in the commissural tracts and backward projection along-
with reduction in FA and increase in RD (Owen et al. 2013; Chang et al. 2015).
DTI metrics from water diffusivity are employed for integrity and connectiv-
ity of white matter. This denotes spatially mean characteristics over voxels.
These consist of many discontinuous, usually components with opposite and
directional nature (Mukherjee et al. 2008). Probabilistic tractography develops
streamlines structure of diffusion fiber tracks considering the directionality
trait for movements of water molecules. Therefore, it frames a map of track
density (TD). These can give more insights about changes in white matter
micro structural (Calamante et al. 2010; 2012a, b). In the human brain, the com-
plex neuron network may be examined by employing graph theory. Here, sub-
cortical and cortical gray matter zones can be denoted by nodes. White matter
interconnective pathways can be represented by links or edges (Bullmore and
Sporns 2009). Edge density imaging (ED) facilitates a structure to denote the
spatial structure, by inserting such connectomic edges in white matter (Owen
et al. 2015; Greene et al. 2017). For full characterization of the neural network
correlation of SPD, one can investigate white matter’s diffusion properties by
employing DTI measures to FA, RD, MD, and AD. Furthermore, probabilistic
tractography can be employed to create maps of ED and TD for comparison of
the white matter connectivity in children who have SPD with other children.

12.2.3 Existing Tools for ASD Diagnosis

ASD diagnostic techniques employ medical experts to perform an evalua-
tion of the person’s development involving various factors (such as self-care,
communication, behavior, social skills). This is known as a Clinical Judgment
approach (CJ) (Wiggins et al. 2014). An Autism Diagnostic Interview (ADI)
and Autism Diagnostic Observation Schedule (ADOS) are frequent evalua-
tion approaches for diagnosis (Lord et al. 1994, 2000). An Autism Diagnostic
Interview (ADI) is an interview-based approach with 93 items. It is usually
performed by a recognized expert in collaboration with the person’s guard-
ian. An ADOS has four parts. Each part is applied to a pre-specified devel-
opment age group, that is implemented along dimensions ranging from
non-verbal persons to verbally fluent persons. People are evaluated by their
reported traits and/or observed traits according to a developmental task set
in each module(Lord et al. 2000).
In addition to the CJ approach, self-evaluation approaches are also defined,
such as Childhood Autism Rating Scale (CARS), Child Behavior Checklist
180 Machine Learning for Healthcare

(CBCL), Autism Spectrum Quotient (AQ) (Baron-Cohen et al. 2001), Autism

Behavior Checklist (ABC), etc. (Thabtah et al. 2018). These approaches are
usually performed by care experts, teachers, parents, or the person them-
selves. These need to be filled out with replies to various questions that
results in many inefficient responses (Thabtah et al. 2018). Therefore, there is
a need to reduce the number of variables in these approaches to make them
more efficient.

12.3 Machine Learning Applied for

Diagnosis of DS, SPD, and ASD
This section identifies machine tools applied in some noteworthy research
on Down’s syndrome, Sensory Processing Disorder, and Autism spectrum
disorder. Table 12.1 shows the machine learning application for disorder
characterization.

12.4 Machine Learning Case Studies of DS, SPD, and ASD

This section presented case studies with a focus on machine learning appli-
cation for DS, SPD, and ASD diagnosis.

12.4.1 Machine Learning (ML) Case Study for DS

Koivu et al. (2018) presented a study with a focus on machine learning
applications for DS diagnosis. Prenatal screening yields a huge amount of
data to predict risk of various disorders based on multiple clinical vari-
ables. Authors applied ML to augment results of the first three months
filtration of DS. ML appears to be an adjustable option to generate better
models for risk evaluation using the existing variables in clinical practice.
This research used two existing datasets with multiple classification algo-
rithms such as deep neural networks, the support vector machine model,
and a risk evaluation business software. Deep neural networks yield a 0.96
area value under the curve, and 78% value of detection rate, with 1% value
of false positive rate for the data. Support vector machine yields 0.95 area
value under the curve, and 61% value of detection rate with 1% value of
false positive rate for the data. Support vector machine yields some poorer
results. Deep neural networks yield greater rate of detection having similar
rates of false positives or same rates of detection and far lower rates of false
positive.
Diagnosis of Disordered Children 181

TABLE 12.1
Machine Learning Applied for Diagnosis of DS, SPD, and ASD
Disorder Authors Machine learning tools Description

DS Grossi et al. Artificial neural network Artificial neural networks have been
2016 investigated to develop model to
predict autism.
Williams et al. logistic regression, NN, For prenatal screening, Williams et al.
1999 classification and proposed logistic regression / linear
regression-tree methods discriminant.
Uzun et al. 2013 Probabilistic classifiers, Authors examined various
machine learning probabilistic classifiers to identify
algorithms, including DS, aimed at decreasing invasive
Decision Tree, SVM, test numbers. They reported SVM
k-Nearest Neighbor (KNN), and multilayer perceptron i.e. feed
and Multilayer Perceptron forward ANN as the best
performing options.
Koivu et al. Deep neural network, Authors applied these algorithms to
2018 Support vector machine enhance identification performance
of DS using the existing clinical
variables and a commercial risk
assessment software.
Neocleous et al. Feed forward neural Neocleous et al. 2018 examined the
2018 network application of a feed forward neural
network for an euploidies
forecasting during the first three
month of pregnancy from data of
Prenatal Non-invasive Testing.
Catic et al. 2018 Neural network (NN)- Authors examined neural network
recurrent, feed forward and (NN) the data of multiple genetic
feedback defects, e.g. DS.
SPD Payabash et al. stepwise penalized logistic These were utilized to detect
2019c, a regression, Naïve Bayes, independent measures to forecast
random forest, support AOR, as possible imaging marker
vector machine (SVM) for AOR.
Tavassoli et al. support vector machine, These were applied to identify
2019 Naïve Bayes, random forest, children with SPD using DTI
neural networks. measures.
ASD Allison et al. Discriminant index (DI) DI decreased variables in the AQ
2012 approach from 50 to 10 variables,
and, in Q-CHAT from 25 parameters
to 10.
Wall et al. Alternating Decision AD Tree decreased the variables/
2012a,b TreeClassifier algorithm items in ADOS-R to 8.
(ADTree) using
WEKA
Duda et al.2016 Logistic Regression, decision Decreased the time required for
tree classifiers Random ADOS and Social Responsiveness
Forest Scale.
Kosmicki et al. Backward step-wide feature Reduced the variables/items in
2015 selection for machine ADOS part 2 and 3.
learning
(Continued)
182 Machine Learning for Healthcare

TABLE 12.1 (CONTINUED)

Machine Learning Applied for Diagnosis of DS, SPD, and ASD
Disorder Authors Machine learning tools Description

Jin et al. 2015 Multi-kernel support vector For ASD-induced defects within 24
machine classification months after birth, authors
proposed a novel classification
framework using the connectivity
measures and diffusion measures
diffusivity.
Payabash et al. support vector machines, These are used to detect ASD children
2019b random forest, Naïve Bayes, with ASD using DTI metrics.
neural networks
Thabtah et al. RIPPER, Decision Tree (C4.5) reduced the items in AQusing
2018, 2017 Va is programmed in Java Variable Analysis (Va) to correlate
and interfaced with WEKA items in three AQ approaches with
3.9.1 to classify ASD. normalised scores of Chi-Square
(CHI) and Information Gain (IG)
approaches.

12.4.2 Machine Learning Case Study of SPD

Payabash et al. (2019c) presented a study with a focus on machine learning
applications for SPD diagnosis. It employed machine learning algorithms to
identify children with SPD using DTI/ tractography metrics. This research
considered 44 SPD0affected children and 41 “typically” developing other
children (DOC) for MRI scanning. It used DTI measures- FA, MD, RD, and
AD. This research also used probabilistic tractography to produce ED and
TD employing DTI maps. Support vector machine, random forest, Naïve
Bayes, and neural networks were used to identify the children with SPD. The
study found accurate rates of classification from measures such as FA, MD,
AD, RD, TD, and ED. This study found that SPD-affected children possess
fewer FA, ED, and TD values, and greater M and RD values in comparison
to DOC, mainly in backside tracts of white matter composed of backside
corona radiata, backside thalamic radiation, backside body, and splenium of
corpus callosum. The average TD of the splenium (p < 0.001) was found as
a single independent measure identifying SPD-affected children from ODC
using stepwise penalized logistic regression. This research found defective
microstructural and connectivity/connectomic integrity in SPD affected
children, mainly in backside tracts of white matter. It also found decreased
TD for splenium of corpus callosum, representing the significant detectable
pattern for identifying children with ASD. It may be applied to design imag-
ing markers of neurodevelopmental defects.

12.4.3 Machine Learning Case Study for ASD

Thabtah et al. (2018) presented a study with a focus on machine learning
applications for ASD diagnosis. ASD is among the most rapidly growing
Diagnosis of Disordered Children 183

developmental disabilities. Most people are not aware ofASD attributes.

Therefore, they do not go for diagnostic services or contact their doctor. Thus,
there is need for a simple tool that employs variables/traits of ASD for affect
families, so that they may increase the probability of seeking expert evalu-
ation. It is critical to detect and treat ASD as early as possible. The machine
learning algorithms used in this research are RIPPER and C4.5. Variable
Analysis (Va) is programmed in Java and interfaced with WEKA 3.9.1 to clas-
sify ASD. In this study, the authors reduced the items in AQ using the Va
to correlate items in three AQ approaches with normalized scores of Chi-
Square (CHI) and Information Gain (IG) approaches. The results shows that
the Va had shown the ability to reduce the features required for screening
approaches from children, adolescents, and adults. This method also main-
tained a good level of accuracy, specificity, and sensitivity.

12.5 Conclusion
Among sustainable development gaols, the social dimension is the most
neglected. More focus and work should be carried out in this area. One
important area of the social dimension that is highly neglected in develop-
ing countries, such as India, is the care of children with disabilities.
This chapter has contributed to the area of developmental care of children/
persons with disorders as follows:

1. It presented examples of methods to identify persons with disabili-

ties using machine learning.
2. It identified the machine learning tools for such research.
3. It also reported on the existing research with novel application of
machine learning in the prenatal and postnatal early diagnosis of
some disabilities such as Down’s syndrome, Autism spectrum disor-
der, and Sensory Processing Disorder to prevent occurrences and/or
to provide early medical care to affected children to facilitate them
to get the right education and training they need to become produc-
tive contributors in the growth of themselves, the nation, and world.

References
Allison C, Auyeung B, Baron-Cohen S (2012) Toward brief “Red flags” for autism
screening: The short Autism Spectrum Quotient and the short quantitative
checklist for autism in toddlers in 1,000 cases and 3,000 controls [corrected].
J. Am. Acad. Child Adolesc. Psychiatry 51(2):202–212.
184 Machine Learning for Healthcare

Baron-Cohen S, Wheelwright S, Skinner R, Martin J, Clubley E (2001) The autism

spectrum quotient (AQ): Evidence from Asperger syndrome/high-function-
ing autism, males and females, scientists and mathematicians. J. Autism Dev.
Disord. 31(1):5–17.
Bruntland G H (1987) World Commission on Environment and Development in: Our
Common Future, United Nations, Oxford University Press, 8–9.
Bullmore E, Sporns O (2009) Complex brain networks: Graph theoretical analysis of
structural and functional systems. Nat. Rev. Neurosci. 10(3):186–198.
Calamante F, Tournier J D, Jackson G D, Connelly A (2010) Track-density imaging
(TDI): Super-resolution white matter imaging using whole-brain track-density
mapping. Neuroimage 53(4):1233–1243.
Calamante F, Tournier J D, Kurniawan N D, Yang Z, Gyengesi E, Galloway G J,
Reutens D C, Connelly A (2012a) Super-resolution track-density imaging stud-
ies of mouse brain: Comparison to histology. Neuroimage 59(1):286–296.
Calamante F, Tournier J D, Smith R E, Connelly A (2012b) A generalised framework
for super-resolution track-weighted imaging. Neuroimage 59(3):2494–2503.
Catic A, Gurbeta L, Kurtovic-Kozaric A, Mehmedbasic S, Badnjevic A (2018)
Application of neural networks for classification of patau, edwards, down,
turner and klinefelter syndrome based on first trimester maternal serum
screening data, ultrasonographic findings and patient demographics. BMC
Med. Genomics 11(1):19.
Centers for Disease Control and Prevention (2014) Prevalence of autism spectrum
disorder among children aged 8 years autism and developmental disabili-
ties monitoring network, 11 sites, United States, 2010. Morb. Mortal. Wkly Rep.
Surveill. Summ. 63(2):1–21.
Chang Y S, Gratiot M, Owen J P, Brandes-Aitken A, Desai S S, Hill S S, Arnett A B,
Harris J, Marco E J, Mukherjee P (2015) White matter microstructure is associ-
ated with auditory and tactile processing in children with and without sensory
processing disorder. Front. Neuroanat. 9:169.
Commission E (2009) Communication from the Commission to the European Parliament
the Council the European Economic and Social committee and the Committee
of the Regions Mainstreaming Sustainable Development into EU Policies: 2009
Review of the European Union strategy for sustainable development brussels.
Cuckle H, Maymon R (2016) Development of prenatal screening: A historical over-
view. Semin. Perinatol. 40(1):12–22. doi:10.1053/j.semperi.2015.11.003.
Duda M, Ma R, Haber N, Wall D (2016) Use of machine learning for behavioral dis-
tinction of autism and ADHD. Transl. Psychiatry 9(6):732.
Elkinton J (2004) Enter the triple bottom line. In: Henriques and J.Richardson(Eds)
The Triple Bottom Line: Does It All Add Up? Assessiong the sustainability of business
andCSSR, London: Earthscan Publications Ltd., 1–16.
Greene C A, Cieslak M, Grafton S T (2017) Effect of different spatial normalization
approaches on tractography and structural brain networks. Netw. Neurosci.
3:1–43.
Grossi E, Veggo F, Narzisi A, Compare A, Muratori F (2016) Pregnancy risk factors in
autism: A pilot study with artificial neural networks. Pediatr. Res. 79(2):339–347.
Jin Y, Wee C Y, Shi F, Thung K H, Ni D, Yap P T, Shen D (2015) Identification of infants
at high-risk for autism spectrum disorder using multi parameter scale white
matter connectivity networks. Hum. Brain Mapp. 36(12):4880–4896.
Diagnosis of Disordered Children 185

Koivua A, Ko Korpim T, Kivel P, Pahikkal T, Sairanen M (2018) Evaluation of machine

learning algorithms for improved risk assessment for Down’s syndrome.
Comput. Biol. Med. 98:1–7.
Kosmicki J, Sochat V, Duda M, Wall D (2015) Searching for a minimal set of behaviors
for autism detection through feature selection-based machine learning. Transl.
Psychiatry 5:514.
Lokesh K S, Jain P K, Sharma A K (2018a) Tactical supply chain planning for Tyre
remanufacturing considering carbon tax policy. Int. J. Adv. Manuf. Technol.
97(1–4):1505–1528. doi:10.1007/s00170-018-1972-03.
Lokesh K S, Jain P K, Sharma A K (2018b) A fuzzy goal programme with carbon tax
for Brownfield tyre remanufacturing supply chain planning. J. Cleaner Prod.
198:1–17. doi:10.1016/j.jclepro.2018.07.005.
Lokesh, K., Jain, P. K. (2016) Tyre- remanufacturing supply chain management prac-
tices: an exploratory study to improve competitiveness and environmental
friendliness. EBSCC 2016 Conference, IIT Kharagpur, India, 12–14 February
2016.
Lokesh K S, Jain P K Sharma A K (2020) A fuzzy goal programme based sustainable
Greenfield supply network design for tyre retreading industry. International
Journal of Advance Manufacturing Technology, 108, 2955–2980.
Lord C, Risi S, Lambrecht L, Cook E H, Leventhal B L, DiLavore P C, Pickles A, Rutter
M (2000) The autism diagnostic observation schedule generic: A standard mea-
sure of social and communication deficits associated with 1the spectrum of
autism. J. Autism Dev. Disord. 30(3):205–223.
Lord C, Rutter M, Le Couteur A (1994) Autism diagnostic interview-revised:
A revised version of a diagnostic interview for caregivers of individu-
als with possible pervasive developmental disorders. J. Autism Dev. Disord.
24(5):659–685.
Mukherjee P, Berman J I, Chung S W, Hess C P, Henry R G (2008) Diffusion ten-
sor MR imaging and fibertractography: Theoretic underpinnings. AJNR Am. J.
Neuroradiol. 29(4):632–641.
Neocleous A C, Syngelaki A, Nicolaides K H, Schizas C N (2018) Two-stage approach
for risk estimation of fetal trisomy 21 and other aneuploidies using compu-
tational intelligence systems. Ultrasound Obstet. Gynecol. Offic. J. Int. Soci.
Ultrasound Obstet. Gynecol. 51(4):503–508.
Owen J P, Chang Y S, Mukherjee P (2015) Edge density imaging: Mapping the ana-
tomic embedding of the structural connectome within the white matter of the
human brain. Neuroimage 109:402–417.
Owen J P, Marco E J, Desai S, Fourie E, Harris J, Hill S S, Arnett A B, Mukherjee P
(2013) Abnormal white matter microstructure in children with sensory pro-
cessing disorders. NeuroImage Clin. 2: 844–853.
Payabvash S, Palacios E, Owen J P, Wang M B, Tavassoli T, Gerdes M R, Brandes
Aitken A, Mukherjee P, Marco E J (2019a) White matter connectome correlates
of auditory over-responsivity: Edge density imaging and machine-learning
classifiers. Front. Integr. Neurosci. 29:10.
Payabvash S, Palacios E M, Owen J P, Wang M B, Tavassoli T, Gerdes M, BrandesAitken
A, Marco E J, Mukherjee P (2019b) Diffusion tensor tractography in children
with sensory processing disorder: Potentials for devising machine learning
classifiers. NeuroImage Clin. 23:1018.
186 Machine Learning for Healthcare

Payabvash S, Palacios E M, Owen J P, Wang M B, Tavassoli T, Gerdes M, Brandes

Aitken A, Cuneo D, Marco E J, Mukherjee P (2019c) White matter connectome
edge density in children with autism Spectrum disorders: Potential imaging
biomarkers using machine-learning models. Brain Connect. 9(2):209–220.
Seuring S (2013) A review of modelling approaches for sustainable supply chain
management. Decis. Support Syst. 54(4):1513–1520.
Tavassoli T, Brandes-Aitken A, Chu R, Porter L, Schoen S, Miller L J, Gerdes M
R, Owen J, Mukherjee P, Marco E J (2019) Sensory over-responsivity: Parent
report, direct assessment measures, and neural architecture. Mol. Autism 10:4.
doi:10.1186/s13229-019-0255-7.
Thabtah F (2017) Autism spectrum disorder screening: Machine learning adapta-
tion and DSM-5 fulfillment. In: Proceedings of the 1st International Conference on
Medical and Health Informatics 2017 Taichung City Taiwan ACM, pp. 1–6.
Thabtah F, Kamalov F, Rajab K (2018) A new computational intelligence approach
to detect autistic features for autism screening. Int. J. Med. Inform. 117:112–124.
Uzun O, Kaya H, Gürgen F, Varol F G (2013) Prenatal risk assessment of trisomy
21 by probabilistic classifiers. In: 21st Signal Processing and Communications
Applications Conference (SIU), Vol. 2013, pp. 1–4. doi:10.1109/SIU.2013.6531604.
Wald N J, Rodeck C, Hackshaw A K, Walters J, Chitty L, Mackinson A M (2003) First
and second trimester antenatal screening for Down’s syndrome: The results
of the serum, urine and ultrasound screening study (suruss). Health Technol.
Assess. 7:1–77 (Winchester, England).
Wall D, Dally R, Luyster R, Jung J, Deluca T (2012a) Use of artificial intelligence to
shorten the behavioural diagnosis of autism. PLoS One 7(8):43855.
Wall D, Kosmiscki J, Deluca T, Harstad L, Fusaro V (2012b) Use of machine learn-
ing to shorten observation-based screening and diagnosis of autism. Transl.
Psychiatry 2:52–58.
Wiggins L, Reynolds A, Rice C, Mody E, Bernal P, Blaskey L, Rosenberg S, Lee L,
Levy S (2014) Using standardized diagnostic instruments to classify children
with autism in the study to explore early development. J. Autism Dev. Disord.
45(5):1271–1280.
Williams C J, Lee S S, Fisher R A, Dickerman L H (1999) A comparison of statistical
methods for prenatal screening for down syndrome. Appl. Stoch. Model Bus.
Ind. 15(2):89–101. doi:1​0.100​2/(SI​CI)15​26402​5(199​904/0​6)15:​2¡89:​:AIDA​SMB36​
6¿3.0​.CO;2​-K.
13
Forecasting Dengue Incidence Rate in Tamil
Nadu Using ARIMA Time Series Model

S. Dhamodharavadhani, R. Rathipriya

CONTENTS
13.1 Introduction................................................................................................. 187
13.2 Literature Review....................................................................................... 190
13.2.1 Findings............................................................................................ 190
13.3 Methods and Materials.............................................................................. 190
13.3.1 Study Area....................................................................................... 190
13.3.2 Snapshot for Dataset....................................................................... 190
13.3.3 Proposed Model.............................................................................. 193
13.3.4 Estimate and Develop the Model................................................. 193
13.4 Results and Discussions............................................................................ 194
13.5 Conclusion................................................................................................... 201
13.6 Acknowledgment........................................................................................ 201
References.............................................................................................................. 201

13.1 Introduction
Mosquitoes are one of India’s toxic insects. They have the ability to carry
and spread disease to humans and this causes millions of deaths every year.
In the year 2015, there 10,683 outbreaks of India Similarly, the worldwide
incidence of dengue has risen 30-fold in the past 30 years, and an increasing
number of countries are reporting their first outbreaks of the disease. The
Aedesa Aegypti mosquito transmits Mosquito-Borne Disease (MBD) such
as chikungunya, dengue, yellow fever, and zika virus to humans. Sustained
and effective mosquito control efforts are necessary to avoid outbreaks of
such diseases (Yong-Su Kwon 2015).
In India, the challenge MBD poses is serious because the increases in
geographic distribution of vectors and MBD have the potential to affect
90% of the population. MBD is mostly an urban public health problem;
however, outbreaks are being increasingly documented in rural areas too
(Dhamodharavadhani and Rathipriya 2016).

187
188 Machine Learning for Healthcare

Geographical Factors of Dengue:

1. Water
2. Housing
3. Climate change
4. Poverty
5. Air Travel
6. Health System

Climate change has had substantial consequences in the global distribution

of MBD. Climate change often impacts dengue transmission, as mosquitoes
grow faster in higher temperatures and bite more often. Researchers have
developed a tool that can forecast the possibility of dengue outbreaks in
different parts of India on the basis of environmental conditions; a devel-
opment that can help to take preventive measures against deadly infection
(Descloux 2012).
A climate-based model of dengue prediction will assist health authorities
in assessing disease intensity in a country or region. On this basis, authori-
ties may have systematic a plan for disease control well in advance and opti-
mize the use of available resources for the same reason (Descloux 2012).
Dengue has been known in India since the 1940s, but the spread of the
disease was previously very limited. The most important MBDs that affect
humans are dengue viruses. According to the World Health Organization
(WHO), dengue is divided into two: types 1 and type 2. Type 1 is a common
form dengue that leads to dengue fever, Dengue Hemorrhagic Fever (DHF)
is type 2. Four types of Dengue Hemorrhagic Fever exist: DHF1, DHF2,
DHF3, and DHF4. Infection of dengue is one of the world’s fastest spreading
MBDs – a viral disease that accounts for nearly 50 million cases annually
(Allard 1998).

• The rate of dengue virus transmission is dangerously high due

to global warming, climate change, rapid urbanization, unsuit-
able sanitation, insufficient public health services, and migratory
populations.
• Regions such as East Mediterranean, Latin America, South East
Asia, and Western Pacific, and Africa are all susceptible to recurrent
dengue fever outbreaks.

Figure 13.1 indicates that in 2017 the dengue cases had been the highest
in a decade. In dengue cases, an increase of more than 300% occurred in
2009, and the total MBD-related death cases in 2017 was the highest in the
last decade. According to data from the National Program for Vector Borne
Disease Control (NVBDCP) and National Health Profile of 2018, dengue
Forecasting Rates of Dengue 189

FIGURE 13.1
Dengue cases and deaths in India.

cases rose to 188,401 in 2017 a than a 300% leap from less than 60,000 cases
in 2009. It is more than a 250% jump compared to the 75,808 events in 2013
(Chiung Ching Ho 2015).
For example, Tamil Nadu has seen Dengue cases rise to 20,945 in 2017 and
Puducherry’s union territory has registered 4,507 dengue cases for the same
year. This means that 2% of the entire population of Puducherry has been
affected by dengue.
Other southern states like Kerala and Karnataka have been badly affected
by this huge dengue outbreak. Kerala showed a dramatic rise from 7,439
cases of dengue in 2016 to 19,638 in 2017. In Karnataka, 16,209 people were
affected by it which translates as 260% higher than in 2016 (Chandran and
Azeez 2015).
As such, successful monitoring and prediction of the incidence rate of MBD
is important in preventing disease spread. Passive, preventive, and reactive
monitoring systems are used to monitor the MBD outbreaks and incidence
rate in Tamilnadu (Dhamodharavadhani and Rathipriya 2020a). Nonetheless,
these programs face problems such as preference for eradication over sur-
veillance, difficulty in interpreting findings, and most importantly, lack of
coordination between MBD eradication units and MBD monitoring units
(Karnaboopathy and Venkatesan 2018).
As a result of the above factors, there is a compelling case for alternative
forms of MBD tracking and forecasting. Therefore, this chapter proposed an
approach based on ARIMA time-series model for forecasting MBD incidence
rate using meteorological data.
190 Machine Learning for Healthcare

13.2 Literature Review
This section provides a state-of-the-art ARIMA models as used in the litera-
ture for predictive analytics. Table 13.1 presents the comparative analysis of
various works.

13.2.1 Findings
From this study, it is clear that efficient and effective dengue forecast tool
with higher accuracy a pressing need in order to control and prevent such
outbreaks all over the world. This study reveals that time series model using
meteorological data has been successful aimed to forecast of specific trans-
missible diseases.

13.3 Methods and Materials

13.3.1 Study Area
In this research, two datasets are taken to predict the dengue incidence rate
using meteorological data in Tamil Nadu. Dataset 1 contains weekly den-
gue cases data for the periods 1990–2009 of Tamil Nadu. Dataset 2 contains
weekly minimum temperatures, maximum temperatures, average tempera-
tures, rainfall, and precipitation for the period of 1990–2009 in Tamil Nadu
District (Yong-Su Kwon 2015)

13.3.2 Snapshot for Dataset

Dataset 1
TABLE 13.1
Comparative Analysis
Technique Description Dataset Recommendation/Prediction
Exponential smoothing STL decomposition forecasts Malaysian Open Data Portal To study the impact of the out-of-sample data
ARIMA trend and seasonal patterns in a on the models and integrate correlation
better way. studies with meteorological data.
(Chiung Ching Ho 2015) Prediction: Dengue fever outbreaks in Malaysia
Trend of morbidity and Tamil Nadu has been witnessing Morbidity and dengue-related (Sahanaa and Mishra 2018)
mortality a fall in the morbidity trend over mortality at Puducherry and Tamil This work had assessed the Dengue in
the past five years, and dengue Nadu, India (2012–2016). Puducherry and Tamil Nadu and proposed
Forecasting Rates of Dengue

deaths since 2015. Nevertheless, the model to reduce and eradicate the dengue
cases of dengue morbidity are outburst
increasing and median mortality Prediction: early detection of the dengue
is rising from 2012 to 2016, outbreak, in fact prediction of dengue
nationally. (Sahanaa and Mishra 2018)
Statistical analysis and Dengue cases in Noumea were Epidemiological data: (Earnest 2012) (Elodie Descloux 2012)
modeling: Time series basically driven by climate January 1971– December 2010 Prediction: Climate-based Dengue Epidemic
analysis, Bivariate during the last forty years. Meteorological data: Models for Understanding and Forecasting
analysis, Multivariate (Descloux 2012) January 1971 to December 2010
analysis. Entomological surveillance data:
since 1997
support vector machine To examine the spatial and Mosquito data collected from 2011 Applied for the efficient control of urban
(SVM), classification temporal variations in the to 2012 at 12 locations, and mosquitoes.
and regression tree frequency of urban mosquitoes environmental data (Yong-Su Prediction: Mosquito Occurrence (Yong-Su
(CART), and random and the relationships with Kwon 2015). Kwon 2015).
forest (RF). meteorological and habitat
conditions such as type of
land use.
(Continued)
191
TABLE 13.1 (CONTINUED)
192

Comparative Analysis
Technique Description Dataset Recommendation/Prediction
Auto Regressive The trend in forecast dengue Dengue cases in Tamil Nadu (Karnaboopathy and Venkatesan 2018)
Integrated Moving cases for the years 2018 to 2025 1997–2017. To avoid the disease from becoming endemic,
Average (ARIMA) shows that there is a stable new interventions with increased intensity of
Model. growth of Dengue cases, which existing interventions and help from the
is of serious concern. international community together with the
WHO are essential in order to stop the
epidemics.
Prediction: Number of cases till December 2025
Seasonal Autoregressive To model the monthly number of Dengue fever (DF) cases in Dhaka Separate modelling approaches for
Integrated Moving dengue fever (DF) cases in Bangladesh monthly data January DF, DHF and DSS would provide better
Average (SARIMA) Dhaka, Bangladesh, 2000 to October 2007 (M.A.H. information to policy-makers and planners
models and forecast the dengue incidence Zamil Choudhurya and Banu, Prediction: forecast for the period.
using time series analysis (Zamil 2008). November 2007 to December 2008.
Choudhurya and Banu 2008).
Machine Learning for Healthcare
Forecasting Rates of Dengue 193

Dataset 2

13.3.3 Proposed Model
Time series analysis Brockwell (2013) may be classified as linear and non-
linear (Dhamodharavadhani Rathipriya2020b). Extremely specific tech-
niques are used for the study of time series ARIMA, such as the Box-Jenkins
multivariate and Holt winter exponential smoothing (single, double, and tri-
ple). ARIMA models are traditional forecasting models that require histori-
cal empirical data as evidence to make predictions (Dhamodharavadhani
and Rathipriya 2019). This model is a simple statistical framework that can be
used as the basis for mathematical models. These three order variables (a, b,
c) describe the process of fitting the ARIMA model to the Box-Jenkins system
(Sahanaa and Mishra 2018).
The Figure 13.2 shows the methodology of time series forecasting using
ARIMA model. The first step of the ARIMA model is preprocessing the data.
Time series data are plotted and its patterns and irregularities are examined.
Next, the outliers and missing values are removed. The second step is to
decompose the data and then to stationary the series and after that to calcu-
late the autocorrelation by choosing the model order based on the process to
fit the ARIMA model, then to evaluate and improve the model. The follow-
ing Table 13.2 describes the steps to fit the ARIMA to a dengue forecasting
model and its mathematical format.

13.3.4 Estimate and Develop the Model

The prediction is iterated in two ways to enhance the forecast. One is to add
the seasonal variable to be removed earlier. Another solution would be to
require (A, B, C) components to be included in the model; refitting the model
on the same results, so that its silent ability is determined by a particular
seasonal trend in the sequence, with the seasonal aspect defined in AR (1)
(Dalinina 2017). The parameters (a, b, c) have also been modified to include
the seasonal variable. Once the same process of analyzing the residual model
has been done, the ACF/PACF plots change the structure as appropriate. For
194 Machine Learning for Healthcare

FIGURE 13.2
Workflow of proposed model.

example, finding the same higher-order evidence is present in auto correla-

tions with lag 7, which indicates a higher-order component may be required.

13.4 Results and Discussions

By plotting the data, the outliers, randomness, or irregularities can be found
as shown in Figure 13.3. In this case, dengue cases show a lot of variations
from one week to another. Despite this, these variations do have a pattern.
For example, the number of dengue cases is less in the winter months and a
higher number dengue cases are observed in the summer months.
In total the number of dengue cases decreased less than 100 on the first
week and to over 400 on the following week as shown in Figure 13.4. These
are alleged outliers that by skewing the statistical summaries may distort the
model. A preprocessing method is used to remove time series outliers. This
Forecasting Rates of Dengue 195

TABLE13.2
Proposed Model Description
Model Mathematical Equations Model parameter
Preprocessing using m = 2k + 1 c = number of terms in model
Moving Average m = series average Y
(MA(c)) k = period point
ARIMA model Rt = c + a1Rb t-1 + a a eb t-a + ¼ Rb is R differenced b times
c = constant
+a1et-1 + a a e a-1 + et et = error
a = alpha
Decomposition xt = T + S + R T = Trend
S = seasonal
xt = T * S * R
R = Random of series
Y = St * Tt * Et xt = x differenced t times
Y = total number of dengue
cases
St = seasonal variable
Tt = pattern and cycle
Et = rest of the error
Stationary Ydt = Yt - Yt -1 d=2
Y = total number of dengue
Yd 2t = Ydt - Ydt -1 = (Yt - Yt -1 ) - ( Yt -1 - Yt - 2 ) cases
t = times point
Yd 2t = Ydt - Ydt -1 = (Yt - Yt -1 ) - ( Yt -1 - Yt - 2 )

å (y - y )(y - k )
Autocorrelations n
, - y,
rk = lag k autocorrelation
t = k +1
t t k = the time lag
rk = n = total number of dengue
å (y - y )
n 2
,
t
t =1 cases
Fitting Dengue Y = total number of dengue
Yˆdt = 0.4551Yt -1 - 0.3496et -1 + E
ARIMA Model cases
t = times point
E = error
original dengue cases
differentiates from order 1

will identify and remove outliers with time series smoothing and decompo-
sition. This process is effective in passing missed values to the array.
The more varied the gap of the moving average, the smoother the original
series develops. For Tamil Nadu’s dengue dataset, the weekly moving aver-
age is considered for smoothing the series.
The Figure 13.5 shows decomposition of data into different components.
They are: data variables represented as weekly data; trend variable is com-
plete pattern series; seasonal variable denotes that fluctuations in the data
related to calendar cycles; and remainder elements consisting of decreasing
or increasing non-seasonal trends.
Figure 13.6 shows that the left panel is stationary, where data values oscil-
late with a constant variance around the average of 1. The right side of the
196 Machine Learning for Healthcare

FIGURE13.3
weekly dengue cases in Tamil Nadu.

FIGURE 13.4
Cleaned data.

plot displays a non-stationary sequence; thus the series mean value will dif-
fer across various time periods. Figure 13.6 shows stationary and non-sta-
tionary series
The meteorological data is non-stationary; the total number of cases of den-
gue varies over time, and shifts in scales. A proper ADF test does not reject
the non-stationary null hypothesis which confirms graphical representation.
In Figure 13.7 ACF shows correlation between a series and its lags. ACF
plots can assist in taking responsibility for MA(b) (Dalinina 2017) order in
Forecasting Rates of Dengue 197

FIGURE 13.5
Decomposition of data. Left panel.

FIGURE 13.6
Stationary and non-stationary series.

addition to recommending the order of variation. PACF denotes that a rela-

tionship between a variable and its lags that could not be explained by previ-
ous lags. It useful to fit the AR(a) order model.
As blue dotted lines, the plots are borders of 95% importance. There are
important meteorological data auto-correlations with many lags, as shown
in the ACF map. It is possible to the transfer effect after the initial lags, as
shown in the PACF point at lags (1,7,7).
198 Machine Learning for Healthcare

FIGURE 13.7
ACF and partial ACF.

FIGURE13.8
ACF and PACF for differenced series.

To continue with the order d = 1 and reassess if further distinction is nec-

essary. The Dickey-Fuller modified research on separated data discards the
non-stationary null hypotheses. To plot Separate sequence with no notice-
able effect, oscillate pattern about 0. This indicates that there is ample dis-
tinction between the terms of order 1 and should be included in the model
(Dalinina, 2017).
Figure 13.8 points at similar lags of the differentiated series that help to
inform our model’s choice of a or b. There are important auto similarities at,
and beyond, Lag 1 and 2.
In Table 13.3 shows a summary of the model. Total dengue cases are rep-
resented as response variables and TMAX, TMIN, TAVG, RAINFALL, PRCP
are represented as predictors.
Figure 13.9 indicates plots for configuration residuals of ACF and PACF.
Parameters of the model’s order and construction are correctly defined, so
there are no major autocorrelations to predict.
Forecasting (ARIMA [1,1,7]) using a fitted model to determine forward
forecast horizon h intervals for full predictions, and using the fitted model
to produce such predictions as shown in Figure 13.10.
The light blue line indicates the fit given by the model in Figure 13.11,
and how the model will work in Tamil Nadu in future dengue cases. One
Forecasting Rates of Dengue 199

TABLE 13.3
Dengue ARIMA Model
Model = total dengue cases ~TMAX+TMIN+TAVG+
RAINFALL+PRCP
Deviance Residuals Min MEDIAN MAX
−50.72 −9.82 419.70
Null Deviance 2146476 on 789 degree of freedom
Residual Deviance 1992815 on 784 degree of freedom
AIC 84

FIGURE 13.9
ACF and PACF for model residuals.

FIGURE 13.10
Forecasts from ARIMA (1,1,7).
200 Machine Learning for Healthcare

FIGURE13.11
Forecast compare to the actual values.

FIGURE13.12
Improving Forecasts from ARIMA.

approach is to substitute a portion of our data as a collection of “hold-outs”,

match the model and then compare the predicted values with the actual
observed values (Dalinina 2017)
The blue line that reflects the projection, though, seems very naive: it fol-
lows a straight line fairly soon, which seems unlikely given the series of past
behavior. Note that the model assumes a seasonal-free series and differen-
tiates the original non-stationary results. In other words, the plotted pre-
dictions are based on the assumption that there will be no other seasonal
variations in the data, and the change in number of cases of dengue from one
year to another is more or less constant in mean and variance.
This forecast may be a simplistic model, but it shows the selection pro-
cess for an ARIMA model and could also serve as a baseline against which
to assess when more complex models are designed. All estimates are given
confidence boundaries for Figure 13.12. 80% trust boundaries are shaded in
darker blue (Dalinina 2017) and 95% in lighter blue. Longer-term predictions
will usually have more volatility as the formula will return to future Y on
Forecasting Rates of Dengue 201

expected values. In this case, this is reflected in the outline of the trust limits,
as they begin to grow with rising horizon. The trust boundaries trend may
point to the need for a more reliable model. Looking at the forecast limits, the
estimated error associated with point predictions is very important.

13.5 Conclusion
In this chapter, the time series proposed ARIMA models are used to predict
the number of dengue cases in Tamil Nadu. The result of the ARIMA model
is more accurate. This model provides significant benefits in predicting den-
gue incidence rate in advance. It may help health officials to take effective
timely preventative measures which saves time, cost, and human lives.

13.6 Acknowledgment
The first author acknowledges the UGC- Special Assistance Programme (SAP)
for the financial support to her research under the UGC-SAP at the level of
DRS-II (Ref.No.F.5-6/2018/DRS-II [SAP-II]), 26 July 2018 in the Department of
Computer Science, Periyar University.

References
Allard, R. (1998). Use of Time-Series Analysis in Infectious Disease Surveillance.
Bulletin of the World Health Organization,76(4), 327–333.
Brockwell, P. D. (2013). Time Series: Theory and Methods. Springer Science & Business
Media.
Chandran, R., Azeez, P. (2015). Outbreak of Dengue in Tamil Nadu, India. Research
Communications. 109(1):171–176.
Chiung Ching Ho, T. C. (2015, November 17–19). Time Series Analysis and Forecasting
of Dengue Using Open Data. In: 4th International Visual Informatics Conference,
IVIC 2015. Bangi, Malaysia: Advances in Visual Informatics.
Dalinina, R. (2017, January 10). Introduction to Forecasting with ARIMA in R.
Retrieved from Oracle Data Science Blog: https​://bl​ogs.o​racle​.com/​datas​cienc​e/
int​roduc​tion-​to-fo​recas​ting-​with-​arima​-in-r​
Dhamodharavadhani, S., Rathipriya, R. (2016). A Pilot Study on Climate Data
Analysis Tools and Software. In: 2016 Online International Conference on Green
Engineering and Technologies (IC-GET). IEEE Xplore Digital Library. Coimbatore.
202 Machine Learning for Healthcare

Dhamodharavadhani, S., Rathipriya, R. (2019). Region-Wise Rainfall Prediction

Using MapReduce-Based Exponential Smoothing Techniques. In: Advances in
Big Data and Cloud Computing Coimbatore. Springer (pp. 229–239).
Dhamodharavadhani, S., Rathipriya, R. (2020a). Enhanced Logistic Regression (ELR)
Model for Big Data. In: F. Garcia Marquez (Ed.), Handbook of Research on Big Data
Clustering and Machine Learning (pp. 152–176). IGI Global.
Dhamodharavadhani, S., Rathipriya, R. (2020b). Variable Selection Method for
Regression Models Using Computational Intelligence Techniques. In: P.
Ganapathi, & D. Shanmugapriya (Eds.), Handbook of Research on Machine and
Deep Learning Applications for Cyber Security, IGI global (pp. 416–436).
Earnest, A. T.-S. (2012). Comparing Statistical Models to Predict Dengue Fever
Notifications. Computational and Mathematical Methods in Medicine.1:1–6.
Elodie Descloux, M. M.-C.-P. (2012). Climate-Based Models for Understanding and
Forecasting Dengue Epidemics. PLOS Neglected Tropical Diseases.6(2)e1470.
Karnaboopathy, R., Venkatesan, D. V. (2018). Forecasting Dengue Incidence in Tamil
Nadu, India: An Time Series Analysis. Journal of Comprehensive Health. 6(2),62–66.
Sahanaa, C., Mishra, A. K. (2018). Trend of Morbidity and Mortality of Dengue in
Tamil Nadu and Puducherry, South India. International Journal of Community
Medicine and Public Health, 5(1)322–325.
Yong-Su Kwon, M.-J. B.-R. (2015). Modeling Occurrence of Urban Mosquitos Based
on Land Use Types and Meteorological Factors in Korea. International Journal of
Environmental Research and Public Health. 12(10), 13131–1314.
Zamil Choudhurya, M. A. H., Banu, S. (2008). Forecasting Dengue Incidence in
Dhaka, Bangladesh: A Time Series Analysis. Dengue Bulletin. 32, 29–37.
Index

AdaBoost, 8 Health Information System Framework, 27

ARIMA, 189 Healthcare Analysis, 131
Artificial Intelligence, 4 Healthcare Professionals, 21
Association Rule, 33 High Dimension, 87, 119
Autism Spectrum Disorder, 176 High Dimension Fractionation, 159
Hospital Operational Management, 18
Biosignals, 69
Boltzmann Entropy, 47 Image Analysis, 31
Brain Cancer, 54 Imaging Data, 20

Cancer, 56 Keras based Deep Learning, 102

Categorical Data, 20
Classification, 1 Light Field, 84, 119
Clustering, 33 Linear regression, 6
Convolutional Network, 103 Liver Cancer, 100, 107
Cutting, 88, 115 Logistic Regression, 42

Data Uncertainty, 2 Machine Learning in Healthcare, 26

Decision Entropy-based Logistic Machine learning, 1, 42, 180
Regression, 47 Medical Data, 19
Decision Making, 41 Medical Information System, 17
Decision Tree, 6 Meningioma, 56
Deep Learning, 27, 100 Metaheuristic Algorithm for
DFL Algorithm, 42 Healthcare, 32
Diagnosis of Disease, 29, 146 Metaheuristic, 32
Diagnosis of Special Children, 175 MIS Data, 21
Dimensions of Big Data, 3 MRI, 160
Disease Identification, 146
Down Syndrome, 176 Neural Network, 30, 103

Electroencephalography Parkinson’s Disease, 30

(EEG), 69 Patient Report Analysis, 129
Electromagnetic Radiation, 55 Pattern Recognition, 1
Electronic Sensors, 22 Pre-and Post-operative Operations, 159
Entropy, 42 Prediction, 1
Predictive Model, 42
Feature Extraction, 71
Filter and Wrapper, 42 Radio Frequency Waves, 55
Forecasting Model, 193 Readmission Prediction, 42

Glioma, 56 Sensory Processing Disorder, 176

Graph, 87 Shannon Entropy, 47
Graph-119 Spatial Channel, 69

203
204 Index

Supervised Learning, 6 Transportation Cost, 35

Support Vector Machine (SVM), 6 Tumor, 86, 125
Tumor Identification, 160
TensorFlow, 108
Textual Data, 20 Unsupervised Learning, 9