TRAINING CERTIFICATE

ABSTRACT
Automated defect detection in medical imaging has become the emergent
field in several medical diagnostic applications. Automated detection of tumor
in MRI is very crucial as it provides information about abnormal tissues which
is necessary for planning treatment. The conventional method for defect
detection in magnetic resonance brain images is human inspection. This
method is impractical due to large amount of data. Hence, trusted and
automatic classification schemes are essential to prevent the death rate of
human. So, automated tumor detection methods are developed as it would
save radiologist time and obtain a tested accuracy. The MRI brain tumor
detection is complicated task due to complexity and variance of tumors. In
this project, we propose the machine learning algorithms to overcome the
drawbacks of traditional classifiers where tumor is detected in brain MRI
using machine learning algorithms. Machine learning and image classifier
can be used to efficiently detect cancer cells in brain through MRI.

i
 INDEX
TABLE OF CONTENTS
PAGE
CHAPTER NO TITLE
NO
ABSTRACT Ⅰ
LIST OF FIGURES Ⅱ
LIST OF TABLES Ⅲ
LIST OF ABBREVATIONS Ⅳ
1 INTRODUCTION 1
2 LITERARY REVIEW 3
3 AIM AND SCOPE OF THE PROJECT 5
3.1 AIM OF THE PROJECT 5
3.2 SCOPE OF THE PROJECT 5
3.2.1 AUTOMATED MEDICAL IMAGING 6
3.2.2 ADVANTAGES 6
3.2.3 DISADVANTAGES 6
4 WORKING THEORY OF THE PROJECT 8
4.1 ARTIFICIAL INTELLIGENCE 8
4.2 BRAIN TUMOR 8
4.3 BASIC OPERATION OF NEURAL
9
NETWORK
4.4 TRANSFER LEARNING 10
4.5 ACTIVATION FUNCTION 11
4.6 CONVOLUTION NEURAL NETWORK 11

ii
5 IMPLEMENTATION AND METHODOLOGY 14
5.1 SOFTWARE REQUIREMENT 14
5.2 HARDWARE REQUIREMENT 14
5.3 DATA SET 14
5.4 PRE-PROCESSING TECHNIQUES 15
5.4.1 DATA AGUMENTATION 15
5.4.2 IMAGE PRE-PROCESSING 16
5.4.3 SEGMENTATION 16
5.4.4 FEATURE EXTRACTION 17
5.4.5 MACHINE LEARNING TRAINING AND
17
TESTING
6 RESULT AND DISCUSSION 20
7 CONCLUSION AND FUTURE WORK 24
7.1 CONCLUSION 24
7.2 FUTURE WORK 24
REFERENCE 25
APPENDIX 27
A. SOURCE CODE 27
B. SNAPSHOTS 35
C.PUPLICATION WITH PLAGARISM
37
REPORT

iii
 LIST OF FIGURES

FIGURE
TITLE PAGE NO
NO
ARCHITECTURE DIAGRAM OF OUR
1 13
MODEL
2 MRI SCAN IMAGE OF BRAIN 15
3 IMAGE AGUMENTATION 16
4 MODEL USING CNN ARCHITETURE 18
5 FLOWCHART OF PROPOSED PROJECT 19
6 TEST ACCURACY RESNET 50 20
7 TEST ACCURACY VGG16 21
8 TEST ACCURACY INCEPTION V3 22

iv
 LIST OF TABLES

S.NO TITLE PAGE NO

1 COMPARISON OF THREE MODELS 22

v
 LIST OF ABBREVATIONS

AI - Artificial Intelligence

ML – Machine Learning

MRI – Magnetic Resonance Imaging

CNN – Convolutional Neural Network

GM – Grey Matter

WM – White Matter

CSF – Cerebro Spinal Fluid

SVM – Support Vector Machine

KNN – K-Nearest Neighbour

RF – Random Forest

NN – Neural network

PCA – Principal Component Analysis

IDM – Inverse Difference Moment

GLCM – Gray Level Co-Occurrence Matrix

vi
 CHAPTER – 1

INTRODUCTION

1.1 Introduction
Brain tumor is one of the most rigorous diseases in the medical science. An
effective and efficient analysis is always a key concern for the radiologist in the
premature phase of tumor growth. Histological grading, based on a stereotactic biopsy
test, is the gold standard and the convention for detecting the grade of a brain tumor.
The biopsy procedure requires the neurosurgeon to drill a small hole into the skull from
which the tissue is collected. There are many risk factors involving the biopsy test,
including bleeding from the tumor and brain causing infection, seizures, severe
migraine, stroke, coma and even death. But the main concern with the stereotactic
biopsy is that it is not 100% accurate which may result in a serious diagnostic error
followed by a wrong clinical management of the disease.

Tumor biopsy being challenging for brain tumor patients, non-invasive imaging
techniques like Magnetic Resonance Imaging (MRI) have been extensively employed
in diagnosing brain tumors. Therefore, development of systems for the detection and
prediction of the grade of tumors based on MRI data has become necessary. But at
first sight of the imaging modality like in Magnetic Resonance Imaging (MRI), the
proper visualization of the tumor cells and its differentiation with its nearby soft tissues
is somewhat difficult task which may be due to the presence of low illumination in
imaging modalities or its large presence of data or several complexity and variance of
tumors-like unstructured shape, viable size and unpredictable locations of the tumor.

Automated defect detection in medical imaging using machine learning has

become the emergent field in several medical diagnostic applications. Its application in
the detection of brain tumor in MRI is very crucial as it provides information about
abnormal tissues which is necessary for planning treatment. Studies in the recent
literature have also reported that automatic computerized detection and diagnosis of
1
the disease, based on medical image analysis, could be a good alternative as it would
save radiologist time and also obtain a tested accuracy. Furthermore, if computer
algorithms can provide robust and quantitative measurements of tumor depiction,
these automated measurements will greatly aid in the clinical management of brain
tumors by freeing physicians from the burden of the manual depiction of tumors

The machine learning based approaches like Deep ConvNets in radiology and other
medical science fields plays an important role to diagnose the disease in much simpler
way as never done before and hence providing a feasible alternative to surgical biopsy
for brain tumors. In this project, we attempted at detecting and classifying the brain
tumor and comparing the results of binary and multi class classification of brain tumor
with and without Transfer Learning (use of pre-trained Keras models like VGG16,
ResNet50 and Inception v3) using Convolutional Neural Network (CNN) architecture.

2
 CHAPTER -2

LITERARY REVIEW

2.1 LITERARY REVIEW

Krizhevsky et al. 2012 achieved state-of-the-art results in image classification
based on transfer learning solutions upon training a large, deep convolutional neural
network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-
2010 contest into the 1000 different classes. On the test data, he achieved top-1 and
top-5 error rates of 37.5% and 17.0% which was considerably better than the previous
state-of-the-art. He also entered a variant of this model in the ILSVRC-2012
competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2%
achieved by the second-best entry. The neural network, which had 60 million
parameters and 650,000 neurons, consisted of five convolutional layers, some of
which were followed by max-pooling layers, and three fully-connected layers with a
final 1000-way SoftMax. To make training faster, he used non-saturating neurons and
a very efficient GPU implementation of the convolution operation. To reduce overfitting
in the fully-connected layers he employed a recently-developed regularization method
called ―dropout‖ that proved to be very effective.

Simonyan& Zisserman 2014 they investigated the effect of the convolutional

network depth on its accuracy in the large-scale image recognition setting. These
findings were the basis of their ImageNet Challenge 2014 submission, where their
team secured the first and the second places in the localization and classification
tracks respectively. Their main contribution was a thorough evaluation of networks of
increasing depth using architecture with very small (3×3) convolution filters, which
shows that a significant improvement on the prior-art configurations can be achieved
by pushing the depth to 16–19 weight layers after training smaller versions of VGG
with less weight layers.

Pan & Yang 2010‗ssurvey focused on categorizing and reviewing the current progress
on transfer learning for classification, regression and clustering problems. In this
3
survey, they discussed the relationship between transfer learning and other related
machine learning techniques such as domain adaptation, multitask learning and
sample selection bias, as well as co-variate shift. They also explored some potential
future issues in transfer learning research. In this survey article, they reviewed several
current trends of transfer learning

Szegedyet al.2015 proposed a deep convolutional neural network architecture

codenamed Inception, which was responsible for setting the new state of the art for
classification and detection in the ImageNet Large-Scale Visual Recognition Challenge
2014(ILSVRC14). The main hallmark of this architecture is the improved utilization of
the computing resources inside the network. This was achieved by a carefully crafted
design that allows for increasing the depth and width of the network while keeping the
computational budget constant. His results seem to yield solid evidence that
approximating the expected optimal sparse structure by readily available dense
building blocks is a viable method for improving neural networks for computer vision.

He et al., 2015 introduced the ResNet, which utilizes ―skip connections‖ and batch
normalization. He presented a residual learning framework to ease the training of
networks that are substantially deeper than those used previously. He explicitly
reformulated the layers as learning residual functions with reference to the layer
inputs, instead of learning unreferenced functions. He provided comprehensive
empirical evidence showing that these residual networks are easier to optimize, and
can gain accuracy from considerably increased depth. On the ImageNet dataset he
evaluated residual nets with a depth of up to 152 layers—8×deeper than VGG nets but
still having lower complexity. An ensemble of these residual nets achieves 3.57% error
on the ImageNet test set. This result won the 1st place on the ILSVRC 2015
classification task. He also presented analysis on CIFAR-10 with 100 and 1000 layers.