BREAST CANCER PREDCTION USNG IN MACHINE LEARNNG

Submitted by

SINEKA M
(Register No:713522CA50)

PROJECT REPORT

Submitted to the

FACULTY OF INFORMATION AND COMMUNICATION ENGINEERING

In partial fulfillment for the award of the degree

Of

MASTER OF COMPUTER APPLICATIONS

in

SNS COLLEGE OF TECHNOLOGY, COIMBATORE-641 035

(AN AUTONOMOUS INSTITUTION)

DEPARTMENT OF COMPUTER APPLICATIONS

NOV/DEC-2023
 BONAFIDE CERTIFICATE

It is certified that this project report “BREAST CANCER PREDICTON USING IN

MACHNE LEARNNG” is the bonafide work of Ms. SINEKA M who carried out of this

project work under my supervision. Certified further, that to the best of my knowledge the work

reported herein does not form part of any other project report or dissertation on the basis of

which a degree or award was conferred on an earlier occasion on this or any other candidate.

_______________________ _______________________

HEAD OF THE DEPARTMENT PROJECT GUIDE

Submitted for the Viva-Voice examination held at SNS COLLEGE OF TECHNOLOGY, held on
_______________

_____________________ ________________________

INTERNAL EXAMINER EXTERNAL EXAMINER

 DECLARATION

I SINEKA M declare that the project work entitled “Breast Cancer Prediction Using in
Machine Learning” submitted to SNS COLLEGE OF TECHNOLOGY (AN
AUTONOMOUS INSTITUTION), Coimbatore of the requirements for the award of Degree
of Master of Computer Applications is a record of original work done by me under the
guidance of Dr.S.SUNDARARAJAN ,Professor and Head of the department of Computer
Applications (PG), SNS COLLEGE OF TECHNOLOGY (AN AUTONOMOUS
INSTITUTION), Coimbatore.

Place: Coimbatore Signature of the

Candidate

Date: SINEKA M
 ACKNOWLEDGEMENT

First of all I solicit my humble thanks to god almighty for being with me, guiding me the right
way throughout my project work.

I express my heart-felt gratitude to the management of SNS College of Technology,

Coimbatore, Dr.S.N.SUBBRAMANIAN, Chairman, Dr.S.RAJALAKSHMI, Correspondent,
Dr.S.NALIN VIMAL KUMAR Technical Director and Dr.V.S.VELUSWAMY, Founder
Trustee for providing me with all sorts of support in completion of this project.

I extend my sincere thanks to Dr.S.CHENTHUR PANDIAN, our Respected Principal, SNS

College of Technology, for the encouragement and facilities offered to complete this project.

I express my heartfelt thanks to Dr.S.SUNDARARAJAN, Professor and Head of the

Department of Master of Applications for this encouragement and relevant instructions, which
made me to complete the project.

I also extend my thanks to my internal guide Dr.S.SUNDARARAJAN, Professor and Head of

the Department, who paved the way for me to complete my project successfully. Her
instruction and presence has guided me a lot for completing this project.

Finally, I thank our parents for their financial and moral support and also our friends who
helped us in completing of our project work successfully

iv
 ABSTRACT

For this project, we are using Deep Convolutional Neural Network (CNN) which is an advanced
neural network. The dataset we use contain images (Histopathological images) from real life
which are for both diseased and normal cases. CNN requires a large amount of data and days of
training on high end systems to give optimum results. To avoid that, we use a process called
transfer learning which make use of pertained model (trained on a dataset of 1 million images).
This helps in getting closer to the optimum model. Further, there will be a training performed
with our own dataset which helps us to create the optimum model for breast cancer detection.

v
 TABLE OF CONTENTS
Page No.
Certificate
Declaration
Abstract
Acknowledgement
Table of contents
List of Figures
List of Tables
List of Abbreviations
1 INTRODUCTION
Company Profle 1
1.1 About Project
1.2 Problem Statement
1.3 Objective
1.4 Scope of the Project
1.5 Methodology

2 LITERATURE SURVEY 5

3 SYSTEM REQUIREMENTS SPECIFICATION 13

3.1 Functional Requirements 14
3.2 Non-Functional Requirements 14
3.3 Hardware Requirements 15
3.4 Software Requirements 15

4 SYSTEM ANALYSIS AND DESIGN 16

4.1 Overview of System Design 17

vi
4.2 Feasibility Study 18
4.3 Use Case Diagram 19
4.4 Class Diagram 20
4.5 Sequence Diagram 21

5 IMPLEMENTATION 22
5.1 VGG16 Architecture 23
5.2 Convolution Layer 26
5.3 Filters 28
5.4 Pooling Layer 32
5.5 Feature Extraction 35
5.6 Random Forest 37

6 RESULTS AND DISCUSSION 39

7 TESTING 45

8 CONCLUSION AND FUTURE SCOPE 50

8.1 Conclusion
8.2 Future Scope
8.3 REFERENCES

vii
 LIST OF FIGURES

Figures Page No.

Fig 4.3 Use Case Diagram 18
Fig 4.4 Class Diagram 19
Fig 4.5 Sequence Diagram for training flow 20
Fig 4.6 Sequence Diagram for Prediction 21
Fig 5.1 VGG16 Structure 23
Fig 5.2 Convolution Layer 25
Fig 5.3 Filter 26
Fig 5.4 Filter Used in Convolution Layer 27
Fig 5.5 Filters Used in Max Pooling Layer 28
Fig 5.6 Pooling Layer 29
Fig 5.7 Feature Extraction 30
Fig 5.8 Random Forest 31
Fig 6.1 Input Image 39
Fig 6.2 Output of vgg16 model 40
Fig6.3 Output of vgg16 model(cont...) 40
Fig 6.4 Accuracy Achieved 41
Fig 6.5 Accuracy Achieved (cont.…) 41
Fig 6.6 Accuracy Graph 42
Fig 6.7 Loss Graph 43
Fig 6.8 Output Shown For Specific Image 44

viii
 LIST OF TABLES

Tables Page No. 1.4

Agile Methodology 3

LIST OF ABBREVIATIONS

CAD - Computer Aided Design

DeCAF - Deep Convolutional Activation Features

FC - Fully Connected

GPU - Graphics Processing Unit

VGG - Visual Geometry Group

ix
 CHAPTER 1

INTRODUCTION

COMPANY PROFLE:

Mobility intelligence software operates at the nexus of data and travel,

revolutionizing the way we navigate our world. It harnesses algorithms and AI to decode
complex traffic patterns, offering real-time insights that optimize routes and streamline
transportation. Within the intricate web of city landscapes, this software becomes a
guiding force, alleviating congestion and promoting efficiency. Businesses and urban
centers benefit from its capabilities, forging a future where commuting becomes effortless
and cities evolve into harmonious, interconnected hubs of mobility and sustainability.

This innovative software redefines the urban narrative, creating pathways that align
seamlessly with the pulse of city life. Its impact extends beyond mere convenience,
fostering a landscape where environmental sustainability intertwines with the ease of
mobility. Through continuous evolution and adaptation, it paves the way for a tomorrow
where transportation is not just efficient but a catalyst for a better world, bridging gaps
and shaping a connected, thriving society.

ABOUT PROJECT:

Cancer is currently a deadly disease rising across the globe. Among the several
existing types of cancer, breast cancer (BC) presents two very concerning
characteristics: It is the most common cancer among women worldwide. It presents a
very high mortality rate when compared to other types of cancer.

Histopathological analysis remains the most widely used method for BC diagnosis
and most of the diagnosis continues being done by pathologists. Applying visual
inspection of histological samples under the microscope, automatic classification of
histopathological images is a research topic that can make BC diagnosis faster and
less prone to errors.

1
The detection of breast cancer is done by the analysis of either mammography or
ultrasound imaging and regular check-ups. Pathologists check the microscopic
elements and tissue structure for detailed analysis.

The approach in which we generally refer to as visual feature descriptors or

handcrafted features, follows a more “traditional” approach, where an evaluation of
the combination of six different feature sets and four base classifiers is conducted, and
the final system is defined by the combination that produces the best results in the
validation set.

The reported results clearly show that the latter can achieve higher recognition rates.
However, the development of such system requires longer training time, some tricks
like random patches to improve performance, and still a lot of expertise from the
developer to tweak the system.

1.1 Relevance of the Project

Deep CNN diagnosis provides a second option for image diagnosis which can
improve the reliability of experts’ decision making. Advanced CNN technology has
achieved great success in natural image classification and it has been used widely in
bio-medical image processing. Digitized tissue histopathology has now become
responsive to the application for computerized image analysis.

1.2 Scope of the Project:

• A Literature Review is a type of Review article. A literature review is a

scholarly paper, which includes the current knowledge including Substantive
findings, as well as theoretical and methodological contributions to a
particular topic.
• Literature Reviews are often associated with academic-oriented literature,
such reviews are found in academic journals and are not to be confused with
book reviews that may also appear in the same publication
• Software requirement Specification is a fundamental document, which forms
the foundation of the software development process. It not only lists the
requirements of a system but also has a description of its major feature.

\
 • Analysis is the process of finding the best solution to the problem. System
analysis is the process by which we learn about the existing problems, define
objects and requirements and evaluates the solutions.

• The chapter covers the implementation aspects of the project, giving details of
the programming language and development environment used. It also gives
an overview of the core modules of the project with their step by step flow.
• Result and conclusion chapter gives the result of the project and the snapshots
of the project, its related information including graphs and other information
of the output.
• Testing chapter gives an overview of the various types of testing incorporated
during the entire duration of the project.
• The summary of the work carried, contributions if any, their utility along with
the scope for further work.

1.3 Objectives

• This project aims to allow using the high-resolution histopathological images

as input to existing CNN, avoiding adaptation of models that can lead more
complex and computationally costly architecture.

• Predicting if the cancer diagnosis is benign or malignant based on several

observations/features.

1.4 Methodology
• Collecting information regarding patients and their microscopic biopsy images.
• Inferring patterns, identifying model parameters and deciding features of
interest.
• Building a record structure showing how patients and pathologists can access
data.
• Deploying the prototype for testing, collecting feedback from users,
maintenance and bug fixing.

Agile Methodology

3
1.5 Problem Statement
• Goal: To build a robust model using machine-learned features and
Convolutional Neural Networks for the detection of breast cancer from
histopathological images.

• Most of these recent works related to BC classification are focused on

WholeSlide Imaging (WSI).

• The broad adoption of WSI and other forms of digital pathology has been
facing obstacles such as the high cost of implementing and operating the
technology, insufficient productivity for high-volume clinical routines,
intrinsic technology-related concerns.

• Deep CNN is used which avoids handling large training data and large
training time by taking output of CNN as input.

CHAPTER 2

LITERATURE SURVEY

2.1 PAPER 1:

Title:” Deep Features for Breast Cancer Histopathological Image Classification”

\
 In this work we presented an investigation of the use of DeCAF features for
breast cancer recognition using the BreaKHis dataset. The large size of the BreaKHis
dataset has given us the opportunity to compare, on the same dataset, CNN trained
from scratch with (DeCAF) features repurposed from another CNN trained on natural
images, which often is not possible with medical image datasets since they are too
small. From the results we can observe that these features are a viable alternative for a
fast creation of image recognition systems using deep learning, and this system can
perform better than systems using visual feature descriptors. Compared with a CNN
trained from scratch, DeCAF features present comparable recognition rates. Note that
training a CNN specifically for the problem requires more complex and slower
training schemes.

This result is important for the design of future classification based systems in
computer-aided diagnosis, since it shows that deep learned features, even if obtained
with a CNN trained on other types of images, are valuable. With this study we make one
more step towards transfer learning for medical image analysis and CAD/CADx
systems, where CNN trained on ImageNet enable the detection of nodules in medical
images.

As future work, one direction is to improve the recognition accuracy of

DeCAF features using patches. Further investigation on the size of the patches, as
well as overlapping patches, can be beneficial to increase the accuracies obtained with
DeCAF features. Another investigation that can produce good results is the
combination of these features with other visual descriptors and task-specific CNNs, to
exploit the complementarity of these approaches. In addition, a better investigation on
feature and classifier selection could also improve performance.

2.2 PAPER 2:

Title: “Breast Cancer Classification using Capsule Network with Pre-processed

Histology Images”

The present work is based on the classification of breast cancer using capsule
net architecture. From this work, it is clear that the performance of the conventional

5
architectures can be improved by data pre-processing and parameter tuning. The
results show that this method can be used as an automated tool to assist doctors in
disease diagnosis, which may lead to higher concentration in the treatment at early
stages rather than diagnosis and can increase the cancer survival rate.

The traditional approach consists of visual descriptors or handcrafted features

which evaluates six different feature sets and base classifier for the system and
outputs combination that produces best results in validation set. Where in CNN
several patches of original images is taken for a CNN architecture. The reported
results show the higher recognition rate compared to traditional method. The aim of
present work is classify breast cancer using deep learning so as it can be used as an
automated tool for the pathologist for diagnosis.

In the proposed method, the histology images are fed as an input to the capsule
network architecture and it consist of input layer, hidden layer and output layer. In
fully connected network every neuron in one layer is connected to every other neuron
to another layer which gives probability distribution. Present work is based on
classification of breast cancer using capsule network and performance can be
improved by improving pre-processing and parameter tuning.

2.3 PAPER 3:

Title: “Breast cancer diagnosis from biopsy images with highly reliable random
subspace classifier ensembles”

In this paper, a reliable classification scheme based on serial fusion of Random

Subspace ensembles has been proposed for the classification of microscopic biopsy
images for breast cancer diagnosis. Rather than simply pursuing classification
accuracy, we emphasized the importance of a reject option in order to minimize the
cost of misclassifications so as to ensure high classification reliability. The proposed
two-stage method used a serial approach where the second classifier ensemble is only
responsible for the patterns rejected by the first classifier ensemble. The first stage

\
ensemble consists of binary SVMs which were trained in parallel, while the second
ensemble comprises MLPs. During classification, the cascade of classifier ensembles
received randomly sampled subsets of features following the Random Subspace
procedure. For both of the ensembles the rejection option was implemented by
relating the consensus degree from majority voting to a confidence measure and
abstaining to classify ambiguous samples if the consensus degree was lower than the
threshold.

The effectiveness of the proposed cascade classification scheme was verified

on a breast cancer biopsy image dataset. The combined feature representation from
LBP texture description, Gray Level Co-occurrence Matrix and Curvelet Transform
exploits the complementary strengths of different feature extractors; the combined
feature was proved efficient with respect to the biopsy image classification task. The
two-stage ensemble cascade classification scheme obtained a high classification
accuracy (99.25 %) and simultaneously guaranteed a high classification reliability
(97.65 %) with a small rejection rate (1.94 %). Moreover, the cascade architecture
provides a mechanism to balance between classification accuracy and rejection rate.
By adjusting the rejection threshold in each ensemble, the classification accuracy and
reliability of the system can be modulated to a certain degree according to the
specification of specific applications. For example, medical diagnosis tasks usually
require high accuracy and reliability, therefore the rejection thresholds in each stage
will be set to a high level in order to guarantee the correctness of the diagnosis.

Although the proposed system has shown promising results with respect to the
biopsy image classification task, there are still some aspects that need to be further
investigated. The benchmark images used in this work were cropped from the original
biopsy scans and only cover the important areas of the scans. However, often it is
difficult to find Regions of Interest (ROIs) that contain the most important tissues in
biopsy scans, more efforts therefore need to be put into detecting ROIs from biopsy
images. In this paper, the parameters for the cascade system, such as ensemble size
and rejection threshold, were decided empirically; this may not have produced the
most satisfactory performance with respect to all application contexts. Therefore,

7
some self-adaptive rules or algorithms for automatically optimizing these parameters
would be desirable.

2.4 PAPER 4:

Title: “Breast Cancer Histopathological Image Classification Using Convolutional

Neural Networks”

In this paper, we have presented a set of experiments conducted on the BreaKHis

dataset using a deep learning approach to avoid handcrafted features. We have shown
that we could use an existing CNN architecture, in our case AlexNet, that has been
designed for classifying color images of objects, and adapt it to the classification of
BC histopathological images.

We have also proposed several strategies for training the CNN architectures,
based on the extraction of patches obtained randomly or by sliding a window
mechanism, that allow to deal with the high-resolution of these textured images
without changing the CNN architectures designed for low resolution images. Our
experimental results obtained on the BreaKHis dataset showed improved accuracy
obtained by CNN when compared to traditional machine learning models trained on
the same dataset but with state-of-the-art texture descriptors. Future work can explore
different CNN architectures and the optimization of the hyperparameters. Also,
strategies to select representative patches in order to improve the accuracy can be
explored.

Classification of histopathology images into distinct histopathology patterns

corresponding to the non-cancerous or cancerous condition of the analysed tissue is
the primary goal. Main challenges of this system is dealing with inherent complexity
of histopathological images. CNN has been widely used to achieve results in different
pattern recognition problems. For images of microscopic and macroscopic textures
CNN is able to surpass traditional textural descriptors.

Besides assessing different CNN architectures, we also use different methods

to deal with high resolution texture images without changing CNN architecture used

\
for low resolution images. The goal is to preserve the original tissue structures and
molecular composition allowing to observe it in a light microscope. CNN has
achieved success in image classification problem including medical image analysis.
CNN consists of multiple trainable stages stacked on top of each other followed by
supervised classifier and feature maps. Feature map is a 2D array storing a color
channel of input image. Output consists of a set array where each feature map
represents a particular feature extracted at location of the input.

2.5 PAPER 5:

Title: “Computer-aided diagnosis of breast cancer based on fine needle biopsy

microscopic images”

In this study, four different clustering algorithms for nuclei segmentation were
compared. The methods were applied in a medical decision support system and were
tested in terms of the classification accuracy on real routine medical data acquired
from the Regional Hospital in Zielona Góra. The main question to be answered was
whether relatively simple but fast methods can be successfully applied in
computeraided diagnosis. The results reaching approximately 96–100% are very
optimistic and provethe approach not only to be fast and relatively easy to implement
but also to provide accurate medical information for a cytologist. More sophisticated
segmentation methods might be disadvantageous considering slight or no resulting
improvement along with the large increase in complexity and computational
requirements. However, the presented methods often fail to appropriately identify
overlapping nuclei. Additionally, if a given image is composed of very few nuclei,
then the clustering might give incorrect results. Given the above, it is important to
provide an adequate number of images (in the paper, the authors used 9 images per
patient) to achieve a good detection accuracy.

The clustering algorithm selection impact on the final classification result was
also verified. Although visual inspection showed some discrepancies in the quality of
the segmentation results, the classification accuracy has not confirmed these
differences.

9
 In the future, this research is going to be directed toward a possible application
of the presented approach to virtual slides (VS). VS are images that have extremely
high resolution (9 gigapixels and more) and contain information on the whole slide.
Moreover, there are plans to improve the accuracy for the three-class problem, where
apart from benign and malignant cases, there is also a fibroadenoma case.
Fibroadenoma is a benign tumor that can have some properties that are similar to a
malignant tumor. However, such a task could require a more sophisticated
classification approach.

2.6 PAPER 6:

Title: “Deep Features for Breast Cancer Histopathological Image Classification”

In this work we presented an investigation of the use of DeCAF features for breast
cancer recognition using the BreaKHis dataset. The large size of the BreaKHis dataset
has given us the opportunity to compare, on the same dataset, CNN trained from
scratch with (DeCAF) features repurposed from another CNN trained on natural
images, which often is not possible with medical image datasets since they are too
small. From the results we can observe that these features are a viable alternative for a
fast creation of image recognition systems using deep learning, and this system can
perform better than systems using visual feature descriptors. Compared with a CNN
trained from scratch, DeCAF features present comparable recognition rates. Note that
training a CNN specifically for the problem requires more complex and slower
training schemes.

This result is important for the design of future classification based systems in
computer-aided diagnosis, since it shows that deep learned features, even if obtained
with a CNN trained on other types of images, are valuable. With this study we make one
more step towards transfer learning for medical image analysis and CAD/CADx
systems, where CNN trained on ImageNet enable the detection of nodules in medical
images.

\
 As future work, one direction is to improve the recognition accuracy of
DeCAF features using patches. Further investigation on the size of the patches, as
well as overlapping patches, can be beneficial to increase the accuracies obtained with
DeCAF features. Another investigation that can produce good results is the
combination of these features with other visual descriptors and task-specific CNNs, to
exploit the complementarity of these approaches. In addition, a better investigation on
feature and classifier selection could also improve performance.

2.7 PAPER 7:

Title: “Deep Learning for Magnification Independent Breast Cancer Histopathology

Image Classification”

In this work, we have proposed a general framework based on CNNs for learning
breast cancer histopathology image features. The proposed framework is independent
from microscopy magnification and faster than previous methods as it requires single
training. Speed and magnification independence properties are achieved without
sacrificing the state-of-the-art performance. Magnification independent models are
scalable, new training images from any magnification level could be utilized and
trained models could easily be tuned (fine-tuning) by introducing new samples.

In this work, we have also proposed a multi-task CNN architecture to predict

both the image magnification level and its benign/malignancy
property simultaneously. The proposed model allows combining image data from many
more resolution levels than four discrete magnification levels. In fact, magnification
level prediction could be formulated as a regression problem which is not limited to a
discrete set of levels. Multi-task prediction requires essentially no additional
computation over single-task prediction. Besides, experimental result shows that
classification performance does not degrade in multi-task network.

For the future work, stain normalization, deeper architectures, and splitting the
network before the last fully-connected layer could be investigated. It would be
interesting to observe task-wise early stopping in multi-task architecture. More
importantly, additional data with increased number of patients should be introduced.

11
We believe CNNs are more promising in breast cancer histopathology image
classification than handcrafted features and the data is the key issue to obtain more
robust models.

CHAPTER 3

SYSTEM REQUIREMENTS SPECIFICATION

Software requirement Specification is a fundamental document, which forms the

foundation of the software development process. It not only lists the requirements of a
system but also has a description of its major feature. An SRS is basically an
organization's understanding (in writing) of a customer or potential client's system
requirements and dependencies at a particular point in time (usually) prior to any
actual design or development work. It's a two-way insurance policy that assures that
both the client and the organization understand the other's requirements from that
perspective at a given point in time.

The SRS also functions as a blueprint for completing a project with as little cost
growth as possible. The SRS is often referred to as the "parent" document because all
subsequent project management documents, such as design specifications, statements
of work, software architecture specifications, testing and validation plans, and
documentation plans, are related to it. It is important to note that an SRS contains
functional and non-functional requirements only; it doesn't offer design suggestions,
possible solutions to technology or business issues, or any other information other
than what the development team understands the customer's system requirements to
be.
The SRS functions as a blueprint for completing a project. The goal of preparing the
SRS document is to:
• Facilitate communication between the customer, analyst, system developers,
maintainers.
• To form a foundation for the design phase.

\
 • Support system testing facilities.
• Controlling the evolution of the system

3.1 Functional Requirement

Functional Requirement defines a function of software system and how system

must behave when presented with specific inputs or conditions. These may include
calculations, data, manipulation and processing and other specific functionality. In
this system following are the functional requirement: -

• Training dataset must be loaded

3.2 Non-functional Requirement

Non-functional requirements are the requirements which are not directly concerned
with the specific function delivered by the system. They specify the criteria that can
be used to judge the operation of a system rather than specific behaviours. They may
relate to emergent system properties. Non-functional requirements for this system are
specified as follows:

• Responsiveness of the system needs to be appropriate since timely retrieval of

sensitive health data is essential.
• The software utilised should be portable so that medical institutions can easily
expand to their inter-connected hospitals, spread across locations.
• Privacy of sensitive data should always be maintained and must not be
misused in any manner.
• Researchers requesting for data must be from authorized sources. Proof of
consent is a responsibility that is borne by the medical institution where the
health data is generated.

3.1 Hardware Requirements

• AWS p2. xlarge instance with configuration: 4 processor cores

• 60GB RAM
• Nvidia K80 (24GB) graphics card

13
 • 20GB hard disk space for training
• Testing can be done on PC: 8GB RAM
• Intel i7 processor
• 10GB hard disk space
3.2 Software Requirement

3.2.1 Google Collab

Google Colab is a free cloud service and now it supports free GPU. Improves Python
programming language coding skills. Not only is this a great tool for improving
coding skills, but it also allows to develop deep learning applications using libraries.
Collaboratory, or “Collab” for short, allows you to write and execute Python in your
browser, with zero configuration required, free access to GPUs and easy sharing.

CHAPTER 4

SYSTEM ANALYSIS AND DESIGN

4.1 Overview

Analysis is the process of finding the best solution to the problem. System
analysis is the process by which we learn about the existing problems, define objects
and requirements and evaluates the solutions. It is the way of thinking about the
organization and the problem it involves, a set of technologies that helps in solving
these problems. Feasibility study plays an important role in system analysis which
gives the target for design and development.

4.2 Feasibility Study

All systems are feasible when provided with unlimited resource and infinite time.
But unfortunately, this condition does not prevail in practical world. So, it is both
necessary and prudent to evaluate the feasibility of the system at the earliest possible
time. Months or years of effort, thousands of rupees and untold professional

\
embarrassment can be averted if an ill-conceived system is recognized early in the
definition phase. Feasibility & risk analysis are related in many ways. If project risk is
great, the feasibility of producing quality software is reduced.

4.2.1 Performance Analysis

Analyse Project Performance is the process of comparing actual project cost

and schedule performance to the performance measurement baseline for the purposes
of analysing the current status of a project.

For the complete functionality of the project work, performance analysis of

various algorithms is included. The algorithms are evaluated in terms of its accuracy,
error, and computational time over a wide range of datasets.

4.2.2 Technical Analysis

System is only beneficial only if it can be turned into information systems that will
meet the organization’s technical requirement. Simply stated this test of feasibility
asks whether the system will work or not when developed & installed, whether there
are any major barriers to implementation. Regarding all these issues in technical
analysis there are several points to focus on: -

Changes to bring in the system: All changes should be in positive direction, there
will be an increased level of efficiency and better customer service.

Required skills: Platforms & tools used in this project are widely used. So, the
skilled manpower is readily available in the industry.

Acceptability: The structure of the system is kept feasible enough so that there
should not be any problem from the user’s point of view.

15
Summary

The main aim of this chapter is to find out whether the system is feasible
enough or not. For these reasons' different kinds of analysis, such as performance
analysis, technical analysis etc. is performed.

4.3 Use-Case Diagram

A use case diagram is usually simple. It does not show the detail of the use
cases:

• It only summarizes some of the relationships between use cases, actors, and
systems.
• It does not show the order in which steps are performed to achieve the goals of
each use case.

The use-case diagram corresponding to the project is depicted in Fig 4.3

 There are five users

• Patient

• Pathologist

• Doctor

• Oncologist

• Breast cancer system

\
Fig 4.3 Use-Case Diagram

4.4 Class Diagram

A UML class diagram is made up of:

• A set of classes and
• A set of relationships between classes

The class diagram corresponding to the project is depicted in Fig 4.4.

The class diagram has the following classes:

• Pathologist: This is user interface class which has function for invoking train
and predict.

17
 • Feature extraction: This class has functions for extracting the features from the
dataset and making it suitable for Random Forest.

• Random Forest: This function predicts whether it is Benign or Malignant by

taking output of the feature extraction as its input.

Fig 4.4 Class Diagram

4.5 Sequence Diagram

• As shown in the sequence diagram for training flow, Pathologist invokes train and
trains a Random Forest model based on the training dataset.

• AS shown in the sequence diagram for Prediction, Random Forest invokes predict
whether it is Benign or Malignant.

The sequence diagram corresponding to the project is depicted in Fig 4.5.

\
Fig 4.5 Sequence diagram for training flow

Fig 4.6 Sequence diagram for prediction

19
 CHAPTER 5

IMPLEMENTATION

The implementation phase of the project is where the detailed design is

actually transformed into working code. Aim of the phase is to translate the design
into a best possible solution in a suitable programming language. This chapter covers
the implementation aspects of the project, giving details of the programming language
and development environment used. It also gives an overview of the core modules of
the project with their step by step flow.
The implementation stage requires the following tasks.

• Careful planning.
• Investigation of the system and constraints.
• Design of methods to achieve the changeover.
• Evaluation of the changeover method.
• Correct decisions regarding selection of the platform
• Appropriate selection of the language for application development

5.1 VGG16 Architecture

• VGG-16 is a convolutional neural network architecture, its name VGG-16
comes from the fact that it has 16 layers.

• Its layers consist of Convolutional layers, Max Pooling layers, Fully

connected layers.

• VGG-16 is used to extract the features from mammograms and selecting

optimal features.

• Convolutional layers consists of filters (to extract significant features) and

padding:

• Filters will be of 2 types: 1. horizontal filters 2. vertical filters.

• Padding: adding extra pixels to retain its quality.

\
Fig 5.1: VGG16 Structure

VGG16 is a convolutional neural network model proposed by K. Simonyan and

A. Zisserman from the University of Oxford in the paper “Very Deep Convolutional
Networks for Large-Scale Image Recognition”. The model achieves 92.7% top-5 test
accuracy in ImageNet, which is a dataset of over 14 million images belonging to 1000
classes. It was one of the famous model submitted to ILSVRC-2014. It makes the
improvement over AlexNet by replacing large kernel-sized filters (11 and 5 in the first
and second convolutional layer, respectively) with multiple 3×3 kernel-sized filters
one after another. VGG16 was trained for weeks and was using NVIDIA Titan Black
GPU’s.

The input to cov1 layer is of fixed size 224 x 224 RGB image. The image is passed
through a stack of convolutional (conv.) layers, where the filters were used with a
very small receptive field: 3×3 (which is the smallest size to capture the notion of
left/right, up/down, centre). In one of the configurations, it also utilizes 1×1
convolution filters, which can be seen as a linear transformation of the input channels

21
(followed by non-linearity). The convolution stride is fixed to 1 pixel; the spatial
padding of conv. layer input is such that the spatial resolution is preserved after
convolution, i.e. the padding is 1-pixel for 3×3 conv. layers. Spatial pooling is carried
out by five max-pooling layers, which follow some of the conv. layers (not all the
conv. layers are followed by max-pooling). Max-pooling is performed over a 2×2pixel
window, with stride 2.

Three Fully-Connected (FC) layers follow a stack of convolutional layers (which has
a different depth in different architectures): the first two have 4096 channels each, the
third performs 1000-way ILSVRC classification and thus contains 1000 channels (one
for each class). The final layer is the soft-max layer. The configuration of the fully
connected layers is the same in all networks.

5.2 Convolutional Layer

The convolutional layer is the core building block of a CNN. Convolutional layers
are the layers where filters are applied to the original image, or to other feature maps
in a deep CNN. This is where most of the user-specified parameters are in the
network. Convolutional Neural Networks (CNN or ConvNet) are complex feed
forward neural networks. CNNs are used for image classification and recognition
because of its high accuracy.

• The process is a 2D convolution on the inputs.

• The “dot products” between weights and inputs are “integrated” across
“channels”.
• Filter weights are shared across receptive fields. The filter has same number of
layers as input volume channels, and output volume has same “depth” as the
number of filters.

\
 Fig 5.2: Convolutional Layer

5.3 Filters

A convolution is how the input is modified by a filter. In convolutional

networks, multiple filters are taken to slice through the image and map them one by
one and learn different portions of an input image. As a result, we see an image where
only dark edges are emphasized.

As shown in the figure 5.3 and 5.4, the reading of the input matrix begins at
the bottom right of image. Next the software selects a smaller matrix there, which is
called a filter. Then the filter produces convolution that is moves along the input

23
image. The filter’s task is to multiply its value by the original pixel values. All these
multiplications are summed up. One number is obtained in the end. After passing the
filter across all positions, a matrix is obtained, but smaller than an input matrix.

Fig 5.3: This is how filters work on RGB channels of a image

The network will consist of several convolutional network mixed with non
linear and pooling layers. When the image passes through one convolution layer, the
output of the first layer becomes the input for the second layer. And this happens with
every further convolutional layer.

\
Fig 5.4: Shows filter used in convolutional layer

The non linear is added after each convolution operation. It has an activation
function, which brings non linear property. Without this property a network would not
be intense and will not be able to model the class label.

The pooling layer follows the non-linear layer. It works with width and height of the
image and perform operation on them. As a result, the image volume is reduced. After
completion of series of convolutional, non-linear and pooling layers, it is fully
connected layer. The layer takes the output information from convolutional networks
to the end of the network results in an N dimensional vector.

5.4 Pooling Layer

pooling layer is another building block of a CNN. Its function is to progressively

reduce the spatial size of the representation to reduce the number of parameters and
computation in the network. Pooling layer operates on each feature map
independently. The most common approach used in pooling is max pooling.

Pooling layers provide an approach to down sampling feature maps by summarizing

the presence of features in patches of the feature map. Two common pooling methods
are average pooling and max pooling that summarize the average presence of a
feature and the most activated presence of a feature respectively.

25
 Fig 5.5: Shows filter used in convolutional layer

As shown in the figure 5.5, consider the max pooling with 2x2 window and
stride 2, maximum value in that stride is 2, in the next stride, the maximum value is 1.
Then the maximum value 3 is followed by 1.

By combining these maximum values, a new matrix is formed, which has reduced in
the size of the presentation to reduce the number of parameters and computation in
the network.

• Convolutional layers provide activation maps.

• Pooling layer applies non-linear down sampling on activation maps.
• Pooling is aggressive (discard info); the trend is to use smaller filter size and
abandon pooling.

The addition of a pooling layer after the convolutional layer is a common pattern used
for ordering layers within a convolutional neural network that may be repeated one or
more times in a given model.

The pooling layer operates upon each feature map separately to create a new set of the
same number of pooled feature maps.

\
 Fig 5.6: Pooling Layer

5.5 Feature Extraction

Feature extraction a type of dimensionality reduction that efficiently represents
interesting parts of an image as a compact feature vector. This approach is useful
when image sizes are large and a reduced feature representation is required to quickly
complete tasks such as image matching and retrieval. Feature extraction describes the
relevant shape information contained in a pattern so that the task of classifying the
pattern is made easy by a formal procedure. In pattern recognition and in image
processing, feature extraction is a special form of dimensionality reduction.

27
Fig 5.7 Feature Extraction

When performing deep learning feature extraction, we treat the pre-trained

network as an arbitrary feature extractor, allowing the input image to propagate
forward, stopping at pre-specified layer, and taking the outputs of that layer as our
features.

As shown in the figure 5.7, when an image is selected, the features are
extracted from the image. Then the information features are selected using which we
can classify whether the tumour is benign or malignant.

5.8 Random Forest

Random forests, otherwise known as the random forest model, is a method for
classification and other tasks. It operates from decision trees and outputs classification
of the individual trees. The random forest is a classification algorithm consisting of
many decisions’ trees. It uses bagging and feature randomness when building each
individual tree to try to create an uncorrelated forest of trees whose prediction by
committee is more accurate than that of any individual tree.

\
 Fig 5.8: Random Forest

5.8.1 Features of Random Forests

▪ It is unexcelled in accuracy among current algorithms.

▪ It runs efficiently on large data bases.
▪ It can handle thousands of input variables without variable deletion.
▪ It gives estimates of what variables are important in the classification.
▪ It generates an internal unbiased estimate of the generalization error as the
forest building progresses.
▪ It has an effective method for estimating missing data and maintains accuracy
when a large proportion of the data are missing.
▪ It has methods for balancing error in class population unbalanced data sets.

29
CODE:
Using VGG16 to diagnose breast cancer detection
import pandas as pd import
numpy as np import os from
google.colab import drive
import pickle
//importing python libraries that will be used later in the program

print(os.getcwd()) drive.mount('/content/gdrive') root_path =

'/content/gdrive/My Drive/project/Brest Cancer/'

Instantiate VGG16

import tensorflow as tf
# Import the VGG16 network architecture
//we use VGG16 network architecture for extracting features from the image

from keras.applications import VGG16; //we are

importing VGG16 from Keras application
VGG16 is a pre-trained model, it already has a network trained with some object
detection.

IMG_WIDTH = 50
IMG_HEIGHT = 50
IMG_DEPTH = 3
BATCH_SIZE = 16
//we initialise dimension of image as 50x50x3 with batch of 16 images

# Instantiate the model with the pre-trained weights (no top)

//then we initialise the model in conv_base

\
conv_base = VGG16(weights='./vgg16_weights_tf_dim_ordering_tf_kernels_notop.h
5',
include_top=False,
input_shape=(IMG_HEIGHT, IMG_WIDTH, IMG_DEPTH))

# Show the architecture conv_base.summary()

//here we can see all the layers present in VGG16

Classification using Random Forest on extracted features

from sklearn.ensemble import RandomForestClassifier

tr = train_features.reshape(train_features.shape[0], -1) clf =

RandomForestClassifier(max_depth=5, n_estimators=5)
//storing RandomForest classifier in clf and setting the depthof forest as 5.

print(tr.shape) //shape
of train features

clf.fit(tr, y_train)
//training the training image features that we are extracting along wuth these
corresponding labels.

(8998, 512)
RandomForestClassifier(bootstrap=True, ccp_alpha=0.0, class_weight=None,
criterion='gini', max_depth=5, max_features='auto',
max_leaf_nodes=None, max_samples=None,
min_impurity_decrease=0.0, min_impurity_split=None,
min_samples_leaf=1, min_samples_split=2,
min_weight_fraction_leaf=0.0, n_estimators=5, n_jobs=None,
oob_score=False, random_state=None, verbose=0, warm_start=False)

31
val = val_features.reshape(val_features.shape[0], -1) // reshaping val_features and
storing it in val.

clf.score(val, y_val)
//to check the accuracy of our previous training, i.e. how accurately this model will
predict if a person has cancer or not.

Classification using neural network

import tensorflow as tf

labs = list(set(y))
//storing all the labels in y as a list in labs.

alt_num_labs_train = np.asarray([labs.index(i) for i in y_train])

alt_num_labs_val = np.asarray([labs.index(i) for i in y_val])

model =
tf.keras.models.Sequential([ tf.keras.layers.Flatten(),
tf.keras.layers.Dense(512, activation=tf.nn.relu),
tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(2,
activation=tf.nn.softmax)
])

adm = tf.keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, decay=0.9, ams

grad=False)
//we are storing Adam optimizer in adm.

model.compile(optimizer=adm,
loss='sparse_categorical_crossentropy',
metrics=['accuracy']) model.fit(train_features,
alt_num_labs_train, epochs=15)

\
// training the model with model.fit
//we are taking epochs as 15, so our model will train for 15 times and tries to reach the
best prediction accuracy.

model.evaluate(val_features, alt_num_labs_val)

import tensorflow as tf

# Define the densely connected classifier

NB_TRAIN_SAMPLES = tr_flat.shape[0]
NB_VALIDATION_SAMPLES = tst_flat.shape[0]
NB_EPOCHS = 50
//we are storing all the flattened features of train features in tr_flat and all flatten
features from val_features in tst_flat.

model = models.Sequential()
model.add(layers.Dense(512, activation='relu', input_dim=(1*1*512)))
model.add(layers.Dense(2, activation=tf.nn.softmax))

model.compile( loss='binary_crossentrop
y',
optimizer=optimizers.RMSprop(lr=2e-4),
metrics=['acc'])
reduce_learning = callbacks.ReduceLROnPlateau(
monitor='val_loss', factor=0.2, patience=2,
verbose=1, mode='auto',
epsilon=0.0001,
cooldown=2, min_lr=0)

eary_stopping = callbacks.EarlyStopping(
monitor='val_loss', min_delta=0,
patience=7, verbose=1, mode='auto')

33
callbacks = [reduce_learning, eary_stopping]

# Train the model history

= model.fit(
tr_flat,
num_labs_train,
epochs=NB_EPOCHS,
validation_data=(tst_flat, num_labs_val),
//training the model

callbacks=callbacks
)
import matplotlib.pyplot as plt

%matplotlib inline acc =

history.history['acc'] val_acc =
history.history['val_acc'] loss =
history.history['loss'] val_loss =
history.history['val_loss'] epochs =
range(1, len(acc) + 1) //here we are
graphically representing the
prediction accuracy during training
and validation.

plt.title('Training and validation accuracy')

plt.plot(epochs, acc, 'red', label='Training acc')
plt.plot(epochs, val_acc, 'blue', label='Validation acc')
plt.legend()

plt.figure()

\
//here we are graphically representing the prediction loss during training and
validation.

plt.title('Training and validation loss') plt.plot(epochs,

loss, 'red', label='Training loss') plt.plot(epochs,
val_loss, 'blue', label='Validation loss')

plt.legend()

plt.show()

Classification on a specific image

#path = r'./Dataset/dataset_IDC/8863/0/8863_idx5_x1201_y1251_class0.png'
path = r'./8863_idx5_x1151_y1101_class1.png' import cv2 image =
cv2.imread(path)
//reading the image

image = image.reshape(1, image.shape[0], image.shape[1], image.shape[2]) _sp_features

= conv_base.predict(np.array(image), batch_size=BATCH_SIZE, verbo se=1)
class_lab = ["No Cancer", "Cancer"] sp_features
= _sp_features.reshape(-1, 512)
//extracting features using conv_base i.e. VGG16 architecture and storing it in
sp_features.

lab = model.predict(sp_features) //defining

labels as no cancer or cancer

pred = [np.where(i == max(i))[0][0] for i in lab]

//we can see from the given input image that the tumor is benign or malignant

print("Diagnosis (predicted):", class_lab[pred[0]])

//printing the prediction, whether the tumor is cancerous or not.
35
 CHAPTER 6

RESULTS AND DISCUSSION

• The overall accuracy for breast cancer diagnosis achieved equal to 86%.
• The dataset showed improved accuracy obtained by CNN when compared to
traditional learning techniques.
• Deep CNN diagnosis provides a second option for image diagnosis which can
improve the reliability of expert decision making.
• The following snapshots and graphs define the results or outputs that we will
get after step by step execution of each proposed protocol.

SCREENSHOTS

Fig 6.1: Input image

• Taking a 50X50 size image as an input image

\
• This image is the most affected part/patch of the original scanned image

Fig 6.2: Output of vgg16 model

Fig 6.3: Output of vgg16 model (cont..)

• This image consists of various layers inside vgg16 architecture

• It shows how the image size is decreasing in each layer i.e., input layer, Convo2D
layer, maxpooling layer

37
 Fig 6.4: Accuracy achieved

Fig 6.5: Accuracy achieved (cont...)

• The above figure consists of accuracy achieved for each of the epoch value
• For different epoch value, the accuracy achieved is different

\
• For epoch=18, the maximum accuracy achieved is 94%

Fig 6.6: Accuracy graph

• The above graph is the accuracy graph obtained from the achieved accuracies

• As we can see, as the training accuracy at its peak, which shows the maximum
result

• Once it becomes constant, the maximum result is taken

39
 Fig 6.7: Loss graph

• The above graph shows loss in training and validation

\
 Fig 6.8: Output shown for a specific image

• This figure shows output obtained for a given input image

• Based on the model prediction, it predict whether the image is cancerous or not

• For the input image of size 50X50, the model predicts the diagnosis as Cancer

41
 CHAPTER 7

TESTING

This chapter gives an overview of the various types of testing incorporated during the
entire duration of the project.

7.1 Unit Testing

Testing of an individual software component or module is termed as Unit Testing. It
is typically done by the programmer and not by testers, as it requires detailed
knowledge of the internal program design and code. It may also require developing
test driver modules or test harnesses.

7.2 Component Testing

Component Testing is mostly performed by developers after the completion of unit

testing. Component Testing involves testing of multiple functionalities as a single
code and its objective is to identify if any defect exists after connecting those multiple
functionalities with each other.

7.3 Integration Testing

Testing of all integrated modules to verify the combined functionality after

integration is termed as Integration Testing. Modules are typically code modules,
individual applications, client and server applications on a network, etc. This type of
testing is especially relevant to client/server and distributed systems.

7.4 System Testing

Under System Testing technique, the entire system is tested as per the requirements.
It is a Black-box type Testing that is based on overall requirement specifications and
covers all the combined parts of a system.

\
7.5 Interface Testing

The objective of this Interface Testing is to validate the interface as per the business
requirement. The expected interface of the application is mentioned in the detailed
design document and interface mock-up screens. Checks if the application correctly
connects to the server.

7.6 Compatibility Testing

Compatibility Testing checks whether the application is compatible with the specified
software and hardware requirements and functions efficiently as expected.

7.7 Performance Testing

Performance testing is used to check for appropriate and efficient performance is

shown by the system as per the requirements. The connection requirements are to be
maintained to ensure efficient performance evaluation.

7.8 Usability Testing

Under Usability Testing, User-friendliness check is done. The application flow is

tested to know if a new user can understand the application easily or not, proper help
is documented if a user gets stuck at any point. Basically, system navigation is
checked in this testing.

CHAPTER 8

CONCLUSTION AND FUTURE WORK

8.1 CONCLUSION

The present work is based on the classification of breast cancer using capsule
net architecture. From this work, it is clear that the performance of the conventional
architectures can be improved by data pre-processing and parameter tuning.

43
 The results show that this method can be used as an automated tool to assist
doctors in disease diagnosis, which may lead to higher concentration in the treatment
at early stages rather than diagnosis and can increase the cancer survival rate.

As it is difficult to detect the breast cancer in early stages, doctors can use
CNN as second opinion. CNN provides accuracy and quality compared to other
methods used to predict the breast cancer tumours as Benign and Malignant. CNN can
be used as a second opinion by the doctors to diagnose the patients.

8.2 FUTURE WORK

• One direction is to improve the recognition accuracy of features using patches,

size and the overlapping of patches.

• stain normalization, deeper architectures, and splitting the network before the
last fully-connected layer could be investigated.

• Additional data with increased number of patients should be introduced.

• A better investigation on feature and classifier selection could also improve

performance.

REFERENCES

• P. Boyle and B. Levin. (2008) “World cancer report” 2008. [Online].

Available:http://www.iarc.fr/en/publications/pdfs-online/wcr/2008/wcr
2008.pdf

• S. R. Lakhani, E. I.O., S. Schnitt, P. Tan, and M. van de Vijver, “WHO

classification of tumours of the breast”, 4th ed. Lyon: WHO Press, 2012.

\
• F. Spanhol, L. S. Oliveira, C. Petitjean, and L. Heutte, “A dataset for breast
cancer histopathological image classification,” IEEE Transactions on
Biomedical Engineering (TBME), vol. 63, pp. 1455–1462, 2015.

• “Breast cancer histopathological image classification using convolutional

neural networks,” in 2016 International Joint Conference on Neural
Networks (IJCNN). IEEE, jul 2016, pp. 2560–2567.

• N. Bayramoglu, J. Kannala, and J. Heikkil¨a, “Deep learning for

magnification independent breast cancer histopathology image
classification,” in 23rd International Conference on Pattern Recognition, vol.
1, December 2016.

• Karpathy, A., Toderici, G., Shetty, S., Leung, T., Sukthankar, R., Fei-Fei, L.:
“Largescale video classification with convolutional neural networks”. In:
Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference
on. pp. 1725–1732. IEEE (2014)

• Krizhevsky,A.,Sutskever,I.,Hinton,G.E.:”Imagenet classification with deep

convolutional neural networks”. In: Advances in neural information
processing systems. pp. 1097–1105 (2012)

45