BREAST CANCER CLASSIFICATION AND

PREDICTION USING HISTOPATHOLOGY

IMAGES
SRI LAKSHMI. H, THIRUSIKA S

ABSTRACT recorded 1,62,468 new registered cases and 87,090 reported

Breast Cancer is ranked the number one cancer among Indian deaths. In 2007, the number of new cases for breast cancer was
females with a rate of 25.8 per 100,000 women and a mortality 12775, while the expected number of new cancer patients in
rate of 12.7 per 100,000 women. It is also one of the most 2018 will be 18235. Statistics show that, in the last decade, the
common causes of cancer worldwide. There have been many number of new cancer disease patients increased every year at
biological and non-biological research in the past and present to an alarming rate.
be able to prematurely detect breast cancer. In this project, we
have taken an approach using Deep Learning Neural Network Breast cancer tumors can be categorized into two
CNN to try to predict whether a histopathology image of the broad scenarios.
human tissue is cancerous or non-cancerous, and if cancerous, (i) Benign (Noncancerous)
what stage does it belong to: in-situ, Invasive Ductal Carcinoma (ii) (ii) Malignant (Cancerous).
Stage 1 (IDC 1), IDC 2, IDC 3 and IDC 4. The results show
higher accuracy than the most state of the art methods in this Identification of the benign and malignant tissues is a very
field. important step for further treatment of cancer. Based on the
penetration of the skin and
damage of the tissue medical photography techniques can be
1.INTRODUCTION
classified into two groups.
The aim for this Project is: (i) Noninvasive.
1)To create an Image Classification Model that can accurately (ii) Invasive.
identify and categorize breast cancer cells from the non-
cancerous images. The benefits of this project are; It helps in improvement in early
2)Implement deep learning neural network Convolutional diagnosis. It helps in reducing healthcare costs, and a more
Neural Networks (CNN) to build the Image Classification beneficial influence on worldwide breast cancer management is
Model. all possible because of the abilities of the machine learning
3)To Predict the stages of the identified cancerous cells. techniques in breast cancer diagnosis.
4)Analyze and Optimize the model and evaluate with
performance metrics. 2.EXISTING WORKS
Breast Cancer Detection and Prevention Using Machine
Histopathology refers to the microscopic examination of tissue
Learning [1] by
in order to study the manifestations of disease. Specifically, in
1 Faculty of Computing, Islamia University of Bahawalpur,
clinical medicine, Histopathology refers to the examination of a
Bahawalpur 63100, Punjab, Pakistan
biopsy or surgical specimen by a pathologist, after the specimen
2 Department of Information Systems, College of Computer
has been processed and histological sections have been placed
and Information Science, King Saud University, Riyadh 11543,
onto glass slides. Histopathology slides, on the other hand,
Saudi Arabia
provide a more comprehensive view of disease and its effect on
3 Department of Information and Communication Engineering,
tissues, since the preparation process preserves the underlying
Yeungnam University, Gyeongsan 38541, Republic of Korea
tissue architecture. As such, some disease characteristics, e.g.,
4 Department of Computer Science, College of Computer and
lymphocytic infiltration of cancer, may be deduced only from a
Information Sciences, King Saud University, Riyadh 11543,
histopathology image. The diagnosis derived from a
Saudi Arabia
histopathology image remains the gold standard in diagnosing
a substantial number of diseases including almost all types of
In their research, the scholars proposed an efficient deep
cancer.
learning model that is capable of recognizing breast cancer in
computerized mammograms of varying densities. Their
Due to the anatomy of the human body, women are more research relied on three distinct modules for feature selection:
vulnerable to cancer than men. Among the different reasons for the removal of low-variance features, univariate feature
breast cancer, age, family history, breast density, obesity, and selection, and recursive feature elimination. The craniocaudally
alcohol intake are reasons for breast cancer. and medial-lateral views of mammograms are incorporated.
They tested it with a large dataset of 3002 merged pictures
Statistics reveal that in the recent past the situation has become gathered from 1501 individuals who had digital mammography
worse. As a case study, shows that breast cancer accounts for performed between February 2007 and May 2015. The
14% of cancers in Indian women. It is reported that with every researchers applied six different categorization models for the
four minutes, an Indian woman is diagnosed with breast cancer. diagnosis of breast cancer, including the random forest (RF),
Breast cancer is on the rise, both in rural and urban India. A decision tree (DT), k-nearest neighbors (KNN), logistic
2018 report of Breast Cancer statistics
regression (LR), support vector classifier (SVC), and linear in two different arrays, where one is to store the images, the
support vector classifier (linear SVC). other is to store its type of class, IDC positive or negative,
Automated Breast Cancer Diagnosis Based on Machine indicated with the numbers 0 and 1.
Learning Algorithms [2] by
1 Habib Dhahri 4.MY WORK

This study was based on genetic programming and machine Firstly, the dataset is imported entirely in my Jupyter notebook.
learning algorithms that aim to construct a system to accurately Then the patches are visualized and some observations are
differentiate between benign and malignant breast tumors. The drawn out.
aim of the study was to optimize the learning algorithm. In this
context, they have applied the genetic programming technique Data Visualization:
to select the best features and perfect parameter values of the
machine learning classifiers. The performance of the proposed What does the image look like? And to what ratio is the IDC
method was based on sensitivity, specificity, precision, positive and IDC negative?
accuracy, and the roc curves.
A single image from the dataset is visualized:
Breast Cancer Survival Prediction by
Sathipati and Ho used an optimized SVM regression to identify
miRNA signatures associated with survival time in patients with
lung adenocarcinoma. They used a novel feature selection
algorithm called IBCGA and these features were then fed into
traditional SVR. Although their custom SVR outperformed
other regression methods, it did not generalize well to unseen
validation data. Another issue with this paper was the size of
datasets.

3.DATASET
Breast cancer can develop at any different part of the breast. The
most common form of breast cancer is Invasive Ductal
Carcinoma (IDC). In order to detect IDC, it is through various
methods such as mammography, ultrasound, biopsy and so on.
Through biopsy, histopathology images are derived. The
Dataset that is going to be used for training and testing for the
image classification model will be the Breast Histopathology
Images dataset. Since the dataset is too large, I have taken 1/8
th part, i. e. 46253 images.

The dataset was originally uploaded on the Gleason Case

website: Fig. 1: A sample of IDC Negative sample i. e. Invasive

http://gleason.case.edu/webdata/jpi-dl Healthy and Non healthy patches are visualized

tutorial/IDC_regular_ps50_idx5.zip then,

The dataset consists of 277,524 50x50 pixel RGB digital image

patches that were derived from 162 H&E-stained breast
histopathology samples. Within these patches, there are 198,
738 IDC negative and 78,786 IDC positive. These images are
small patches that were extracted from digital images of breast
tissue samples. The breast tissue contains many cells but only
some of them are cancerous. Patches that are labeled "1" contain
cells that are characteristic of invasive ductal carcinoma. For
more information about the data, see

https://www.ncbi.nlm.nih.gov/pubmed/27563488
Fig. 2: Healthy Patches
and
http://spie.org/Publications/Proceedings/Paper/10.1117/12.204
3872.

There is a compilation of all the images into one folder which

is named IDC_regular_ps50_idx5. For this project, the data
from IDC_regular_ps50_idx5 folder will be directly extracted.
It is a folder full of folders that is named after the patients' id,
which also consists of the IDC positive and IDC negative
photos. There are 279 patients, and in each file contains images
of IDC positive and negative. The way the images are stored is
 Fig. 3: Cancerous Patches plotted in addition to the Confusion Matrix.
Observations:
 There might be a chance that not all the images are
50x50pixels.
 Comparing the Healthy Patches and the Cancer
Patches, the Cancer patches seems to have more
Purple-ish look to it.

To avoid any issues during training, I have resized all images to

follow the 50x50 size to ensure fairness. After resizing,
normalization, shuffling and splitting the data have been carried
out.

The Model used for this project is a custom Convolutional

Neural Network model. The Optimizer used for this model is
Adam and the evaluation metrics is Accuracy and Confusion
Matrix.
Fig. 4: Accuracy Curve for Model 1
Application or Use of CNN in Breast Cancer Diagnosis:

CNNs and AI can improve medical image quality by enhancing

low-contrast features, reducing noise, removing artifacts, and
optimizing image registration. They also assist in image,
segmentation, and ROI detection, enabling precise analysis and
diagnosis of anatomical structures or lesions. AI algorithms can
adjust image contrast, brightness, and intensity levels and apply
contrast-limited adaptive histogram equalization (CLAHE)
techniques to improve image quality. Additionally, CNNs
recognize and remove common imaging artifacts, ensuring
accurate interpretation. AI algorithms optimize image
alignment, while segmentation and ROI detection enable
precise analysis and diagnosis of specific areas. CNNs is also
used for super-resolution imaging, enhancing image resolution
and quality beyond the original acquisition. AI-driven super-
resolution techniques use Deep learning models to generate
high-resolution images from low-resolution inputs, providing
enhanced detail and diagnostic Fig. 5: Loss Curve for Model 1
information.
To further increase the accuracy of the model, another dense
CLASSIFICATION OF CANCEROUS AND NON
layer was added and early stopping was used. The Model 2 had
CANCEROUS IMAGES
4 Convolutional Layers, 1 Flattening layer and 3 Dense Layers.
Custom CNN Model for classifying between Cancerous and The number of Epochs was 25 with a batch size of 256. The loss
Noncancerous: function used here is Binary Cross Entropy.

My Custom CNN has 4 Convolution and Max Pooling layers, 1 The improved Model’s Accuracy has been increased from 85 to
Flattening Layer and 2 Dense Layers. The Conv2D layers apply 87 percent.
convolutional operations with different filter sizes to capture Accuracy: 0.87 Precision: 0.76
image features. MaxPooling2D layers down sample the spatial Recall: 0.70 F1-score: 0.73
dimensions to reduce computational complexity. The flattening
layer flattens the output from the previous layer into a 1D array.
The Dense layers are responsible for combining the extracted
features and making the final classification. The first layer uses
Relu Activation function and the last layer uses the Softmax
activation function to produce probability scores for each class,
cancerous and non cancerous.

The number of Epochs was 25 with a batch size of 75. The loss
function used here is Binary Cross Entropy.

The Model has an Accuracy of 85 percent.

Accuracy: 0.85
Precision: 0.73
Recall: 0.66
F1-score: 0.70

The Model Accuracy Curve and the Loss curve has been
Fig. 6: Accuracy Curve for Model 2
 Linear Regression Model

The Linear Regression Model was trained and tested

with the same training and testing sets as of the above SVM
Model. The accuracy was very low and with Mean absolute
Error of 1.6497409805180474.

Custom CNN Model

My Custom CNN has 4 Convolution and Max Pooling layers, 1

Flattening Layer and 3 Dense Layers. The number of Epochs
was 2 with a batch size of 32. The loss function used here is
Sparse categorical cross entropy.
The Model had an accuracy of about 19 percent.
Then to further Generalize and improve the accuracy of the
model, Data augmentation technique was used and the model
Fig. 7: Loss Curve for Model 2 was trained again.

Data augmentation is a technique commonly used in computer

CLASSIFICATION OF STAGES vision tasks to artificially increase the size of the training dataset
For the Classification of stages, the RGB Intensity of the by applying various transformations to the existing images. This
patches were analyzed and based on the intensity of the Red and helps improve the generalization of the model and reduces the
Blue channels, which constitutes the intensities for the purple- risk of overfitting.
ish color of the patches.

Fig. 8: Accuracy Curve for this Model

For Cancerous:
Average Color (BGR):
[169.7552 130.784 187.0252]
RGB Ratios:
[0.66570667 0.51287843 0.73343216]

Support Vector Machine for the classification of stages

A Support Vector Machine (SVM) is a discriminative classifier

formally defined by a separating hyperplane. In other words,
given labeled training data (supervised learning), the algorithm
outputs an optimal hyperplane which categorizes new
examples. In two-dimensional space this hyperplane is a line Fig. 9: Loss Curve for this Model
dividing a plane in two parts where in each class lay in either
side. The SVM Model was used initially for classifying the Accuracy: 0.19
stages. The Cancerous Data which had 11,708 images were split Class: In-Situ,
in the ratio of 80% for training and 20% for testing. Precision: 0.00, Recall: 0.00, F1-score: 0.00
The accuracy of this model initially was 19% and upon Class: idc 1,
improvement, the accuracy was increased to 0.203245089666 Precision: 0.19, Recall: 1.00, F1-score: 0.32
951which is 20% percent. Class: idc 2,
Precision: 0.00, Recall: 0.00, F1-score: 0.00
Class: idc 3, Table 1: Classification as Cancerous and Non
Precision: 0.00, Recall: 0.00, F1-score: 0.00 cancerous
Class: idc 4,
Model Accuracy Precision Recall F1 score
Precision: 0.00, Recall: 0.00, F1-score: 0.00
CNN 0.85 0.73 0.66 0.70
Lastly, the model was tested with an image and the predicted Model 1
class was obtained and the image was visualized.
CNN 0.87 0.76 0.70 0.73
Model 2

Table 2: Classification of Cancerous Patches into

Different Stages
Model Accuracy
SVM 0.20324508966695132
Regression Model 0.20324508966695132
CNN Model 0.19214347004890442

9.REFERENCES
[1]Khalid, Arslan, Arif Mehmood, Amerah Alabrah, Bader
Fahad Alkhamees, Farhan Amin, Hussain AlSalman, and Gyu
Sang Choi. 2023. "Breast Cancer Detection and Prevention
Using Machine Learning" Diagnostics 13, no. 19: 3113.
https://doi.org/10.3390/diagnostics13193113

[2]Habib Dhahri, Eslam Al Maghayreh, Awais Mahmood, Wail

Elkilani, Mohammed Faisal Nagi, "Automated Breast Cancer
Diagnosis Based on Machine Learning Algorithms", Journal of
Healthcare Engineering, vol. 2019, Article ID 4253641, 11
pages, 2019. https://doi.org/10.1155/2019/4253641
Fig. 10: A Cancerous patch predicted as IDC Stage 1
[3]Rakhlin, A., Shvets, A., Iglovikov, V. and Kalinin, A.A.,
5.FUTURE WORK 2018. Deep Convolutional Neural Networks for Breast Cancer
Histology Image Analysis. arXiv preprint arXiv:1802.00752
1)I aim to try various machine learning techniques and a
standardized dataset to further find out the stage the patch is in [4]ICIAR 2015 Grand Challenge on Breast Cancer
and predict the duration it will take to reach the next stage. Histology Images.
6.COMPARISION WITH OTHER WORKS [6]Kerketta Z H, Kujur A, Kumari N, et al. (June 15, 2023) A
In the work of Rakhlin et al [3], a Deep Learning CNN approach Cross-Sectional Study on the Epidemiology of Newly
was used to classify breast cancer histology images. The Diagnosed Breast Cancer Patients
accuracy achieved was 86% by this process. In my project work, Attending Tertiary Care Hospitals in a Tribal Preponderant State
we get an 87% accuracy on the classification, which is an of India: Regression Analysis. Cureus 15(6): e40489. DOI
improvement. 10.7759/cureus.40489

7.RESULTS
On assessing the performance of the models of classifying the
cancerous and non cancerous images on the test set comprising
of images from each class, the two models achieves a
remarkable accuracy of 85 and 87 percent respectively.
On assessing the performance of the three models SVM, Linear
Regression Model and the CNN Model, the three models give a
low accuracy of around 19-20 precent. This can be further
improved by using Data Augmentation techniques, using
various advanced machine learning models and incorporating
several other important factors like tumor size, tissue hardness
etc. to high quality standard histopathology image dataset.

8.CONCLUSION
These are the final results obtained from the models. Each
model is curated carefully with modifications and
improvements at each line of code in order to obtain the
desired result.