BRAIN TUMOR DETECTION AND CLASSIFICATION USING

DEEP LEARNING
By

SRIJON BARUA
ID: 201004912

Department of Computer Science & Engineering

School of Science, Engineering & Technology

Supervised By
GOLAM MOKTADER DAIYAN
Assistant professor, SoSET
East Delta University

In partial fulfillment of the requirement for the degree of Bachelor of Science

in Computer Science and Engineering

September, 2024

East Delta University, Noman Society, East Nasirabad, Chittagong-4209

 BRAIN TUMOR DETECTION AND CLASSIFICATION USING
DEEP LEARNING

This thesis is submitted in partial fulfillment of the requirement for the degree of Bachelor of
Science in Computer Science & Engineering.

By

SRIJON BARUA
ID : 201004912

Supervised By
GOLAM MOKTADER DAIYAN
Assistant Professor, SoSET
East Delta University

Department of Computer Science & Engineering

School of Science, Engineering & Technology
East Delta University, Noman Society, East Nasirabad, Chittagong-4209
The undergraduate thesis titled Brain Tumor Detection and Classification Using Deep
Learning submitted by Srijon Barua, 201004912 has been accepted as satisfactory in
fulfillment of the requirement for the degree of Bachelor of Science (B.Sc.) in Computer
Science and Engineering (CSE) to be awarded by East Delta University.

Board of Examiners

Dr Mohammad Mahbubur Rahman Chairman

Assistant Professor
School of Science, Engineering and Technology
East Delta University

Arifa Sultana Zarna Member

Assistant Professor
School of Science, Engineering and Technology
East Delta University

Saraf Anika Member

Lecturer
School of Science, Engineering and Technology
East Delta University

Golam Moktader Daiyan Supervisor

Assistant Professor
School of Science, Engineering and Technology
East Delta University

Ⅰ
 DECLARATION

It is hereby declared that we have produced the work presented in this thesis, during the
scheduled period of study. We also declare that we have not taken any material from any source
except referred to wherever due that amount of plagiarism is within acceptable range.

Date -
Srijon Barua
201004912

Ⅱ
 ACKNOWLEDGEMENTS

First and foremost, I would like to express my profound gratitude to Almighty God for his
continuous blessings, wisdom, and strength that enabled me to complete this thesis.

I would like to extend my heartfelt appreciation to my thesis supervisor, Golam Moktader

Daiyan, for his invaluable guidance, constant encouragement, and expert insights throughout the
research process. His unwavering support and dedication have been crucial to the successful
completion of this work. His mentorship has not only enriched this thesis but also fostered my
academic and personal growth.

I am also deeply grateful to the members of my thesis committee for their constructive feedback,
time, and support. Their insightful comments and suggestions have greatly contributed to the
development of this thesis.

III
 TABLE OF CONTENTS

DECLARATION II

ACKNOWLEDGMENTS III

TABLE OF CONTENTS IV

LIST OF FIGURES VI

LIST OF TABLES VIII

LIST OF ABBREVIATIONS IX

ABSTRACT X

CHAPTER 1
1

INTRODUCTION 1

1.1 Research Background 1

1.2 Motivation 2

1.3 Research Objectives 3

1.4 Thesis Organization 4

CHAPTER 2 5

LITERATURE REVIEW 5

CHAPTER 3 8

RESEARCH METHODOLOGY 8

3.1 Methodology 8

3.2 Dataset 9
IV
3.3 Data Preprocessing 10

3.3.1 Data Splitting 10

3.3.2 Image Augmentation 11

3.3.3 Image Resizing 12

3.3.4 Image Normalization 12

3.4 MobileNetV2 Architecture 13

3.5 InceptionV3 Architecture 16

3.6 ResNet 101 Architecture 18

3.7 VGG19 Architecture 23

3.8 EfficientNetB5 Architecture 26

CHAPTER 4 33

RESULTS AND DISCUSSION 33

4.1 Performance Evaluation 33

4.2 Experimental Setup 34

4.3 Hyperparameters 34

4.4 Experiment with MobileNetV2 38

4.5 Experiment with InceptionV3 40

4.6 Experiment with ResNet101 42

4.7 Comparison of Models 44

4.8 Discussion. 47

CHAPTER 5 49

CONCLUSION AND FUTURE WORK 50

References 51
Ⅴ
 LIST of FIGURES

Figure 1: Methodology of all model 8

Figure 2: Classes of dataset 9

Figure 3: Dataset Split 11

Figure 4: Depthwise Separable Convolutions of MobileNetV2 13

Figure 5: Inverted Residual Block of MobileNetV2 14

Figure 6: Leaner bottleneck of MobileNetV2 14

Figure 7: MobileNetV2 Model Architecture 15

Figure 8: InceptionV3 Architecture 16

Figure 9: InceptionV1 Model Architecture 16

Figure 10: Resnet101 Architecture 18

Figure 11: Convolution layer of ResNet101 21

Figure 12 : Bottleneck block of ResNet101 21

Figure 13 : Fully connected layer and softmax 22

Figure 14 : VGG19 Architecture 23

Figure 15 : EfficientNetB5 Architecture 26

Figure 16 : Compound scaling of EfficientNetB5 27

Figure 17 : Mobile Inverted Bottleneck Convolution (MBConv) 28

Figure 18 : MobileNetV2 accuracy and loss curves 39

Figure 19 : Confusion matrix of MobileNetV2 40

Figure 20 : InceptionV3 accuracy and loss curves 41

Ⅵ
Figure 21 : Confusion matrix of InceptionV3 42

Figure 22 : ResNet101 accuracy and loss curves 43

Figure 23 : Confusion matrix of ResNet101 44

Figure 24 : VGG19 accuracy and loss curves 46

Figure 25 : VGG16 accuracy and loss curves 46

Figure 26 : EfficientNetB5 accuracy and loss curves 46

Figure 27: MobileNet accuracy and loss curves 47

Figure 28: Model Comparison bar chart 47

Ⅶ
 LIST of TABLES

Table 1: Data distribution 10

Table 2:Key components of ResNet101 19

Table 3:Vgg19 Layer structure 25

Table 4: EfficientB5 structure 29

Table 5: Resizing technique 34

Table 6: Augmentation technique 35

Table 7: The parameters passed to MobileNetV2 36

Table 8: The parameters passed to InceptionV3 36

Table 9: The parameters passed to ResNet101 37

Table 10: The parameter passed to VGG19 37

Table 11: Additional layers of all models 38

Table 12: Result of MobileNetV2 38

Table 13 : Result of InceptionV3 40

Table 14 : Result of ResNet101 42

Table 15 : Model comparison 45

Ⅷ
 LIST of ABBREVIATIONS

CNN : Convolutional Neural Network

DL : Deep Learning

ReLu : Rectified Linear Network

GPU : Graphics Processing Unit

RAM : Random Access Memory

GB : Giga Byte

MLP : Multi-Layer Perceptron

Ⅸ
 ABSTRACT

Brain tumors are among the most complex and deadly forms of cancer, presenting substantial
challenges in diagnosis, treatment, and prognosis. This thesis primarily focuses on gliomas, the
most common and aggressive type of brain tumor, delving into the molecular and genetic
mechanisms responsible for their formation, progression, and treatment resistance. Key
biomarkers and signaling pathways are explored, shedding light on how they influence tumor
behavior and therapeutic outcomes. To enhance diagnostic accuracy, this study applies
state-of-the-art deep learning models to automate the detection and classification of brain tumors
using MRI scans. Several CNN architectures—MobileNet V2, Inception V3, ResNet101,
VGG16, and VGG19—were trained and evaluated on medical imaging datasets. A detailed
comparative analysis assesses these models based on computational efficiency, accuracy, and
clinical suitability. The results demonstrate that these models can accurately identify tumor types,
offering significant potential to streamline diagnosis and support clinicians in decision-making.
In addition to machine learning approaches, the thesis critically examines both conventional and
emerging treatment strategies. Standard methods, including surgery, radiotherapy, and
chemotherapy, are discussed, along with their limitations. Moreover, emerging therapies like
immunotherapy and molecularly targeted treatments are explored, focusing on their potential to
improve outcomes in resistant tumors. The study also addresses major challenges, such as the
blood-brain barrier restricting drug delivery and the role of the tumor microenvironment in
fostering resistance to therapy. By integrating deep learning, experimental research, and clinical
data analysis, this thesis aims to contribute to a deeper understanding of brain tumor biology and
enhance diagnostic precision. Ultimately, the insights gained through this research can lead to
more effective treatment strategies, helping to improve patient outcomes and quality of life

Keywords : Brain Tumor Detection, Convolutional Neural Network, Neuroimaging, Magnetic

resonance imaging, Feature extrusion, Deep learning.

Ⅹ
Tab 2