A

Presentation On Project Stage I

"Brain Tumor Classification of MRI Images Using Deep Convolutional
Neural Network”

Presented By
Ms.Pooja Devidas Bhoite(Seat No:-1273)

Guided by
Dr.Y.V.Parkale

SHIVNAGAR VIDYA PRASARAK MANDAL'S

COLLEGE OF ENGINEERING, MALEGAON (BK)-BARAMATI.
Savitribai Phule Pune University, Pune
Academic Year 2024-25 Sem-III
Outline
• ABSTRACT

• INTRODUCTION

• LITERATURE REVIEW

• PROBLEM STATEMENT

• PROPOSED SYSTEM

• SOFTWARE AND HARDWARE SPECIFICATIONS

• CONCLUSION

• REFERENCES
 ABSTRACT

Manual tumor diagnosis from magnetic resonance images (MRIs) is a

time-consuming procedure that may lead to human errors and may

lead to false detection and classification of the tumor type. Therefore,

to automatize the complex medical processes, a deep learning

framework is proposed for brain tumor classification to ease the task of

doctors for medical diagnosis. Publicly available datasets such as

Kaggle and Brats are used for the analysis of brain images. The

proposed model is implemented on three pre-trained Deep Convolution

Neural Network architectures (DCNN) such as AlexNet, VGG16, and

ResNet50. These architectures are the transfer learning methods used

to extract the features from the pre trained DCNN architecture, and the

extracted features are classified by using the Support Vector Machine

(SVM) classifier. Data augmentation methods are applied on Magnetic

INTRODUCTION
 The development of unusual cells in the human brain
is called a brain tumor.
 The Brain tumor is categorized into two types,
primary and secondary brain tumor.
 The Primary brain tumor mainly originates from the
brain or any part of the brain without infecting the
other parts of the body. Malignant and benign are the
most common types of tumors.
 The Secondary brain tumor does not directly
originate in the brain, but the tumor gets spread from
different parts of the body. Malignant tumors are
mostly considered secondary type tumors.
 A benign tumor can be further classified into meningiomas and
gliomas; these are regarded as low-grade tumors. A malignant tumor
is a high-grade tumor, classified into glioblastoma and astrocytoma.
 The imaging practices like Magnetic Resonance Imaging (MRI),
Computed Tomography (CT), are essentially utilized in
investigating the brain images.
 The standard type of brain tumor resulting in adults is Gliomas,
which can be identified with the help of magnetic resonance (MR)
images .
 Early detection of tumor plays an essential role in the treatment
process. The radiologist uses classification methods to categorize
the MR image as normal or abnormal.
 Manual classification is an expensive and time-taking
assignment ,In addition, they can get different conclusions from
various observers.
 Instead of developing a new model, deep transfer learning adapts an
existing deep model that has already proven its effectiveness.
 Transfer learning, leveraging pre-trained models, offers efficient
feature extraction from MRI images.
 There are more than 120 different types of brain tumors,
lesions and cysts, which are differentiated by where they
occur and what kinds of cells they are made of.
 Glioma is a common type of tumor originating in the
brain, but it can sometimes be found in the spinal cord.
About 33% of all brain tumors are gliomas. These
tumors arise from the glial cells that surround and
support neurons.
Gliomas
Gliomas are brain tumors starting in the
glial cells.
Gliomas can be low grade (slow growing) or
high grade (fast growing).
Physicians use the grade of a brain tumor
based on gliomas to decide which treatment
a patient needs. The condition of the tumor
is of vital importance for the treatment
 PROBLEM STATMENT
 The motivation behind this research thesis is to address the challenges faced in accurately

identifying and analyzing brain tumors using traditional systems in hospitals. Traditional

image processing tools often fall short in correctly identifying tumor types and segmenting

specific regions of abnormal tissue, especially in hard-to-reach areas of the brain with

varying edema and tumor levels. While MRI is a non-invasive and effective method for

visualizing soft tissue anomalies, such as brain lesions, the manual analysis of MRI films

by radiologists can be time-consuming and labor-intensive.

 This study aims to apply modern image processing and computer-aided techniques to

automatically identify and quantify anomalies in MRI brain images. By leveraging

advanced methods like transfer learning and soft computing, the goal is to achieve accurate

classification of brain tumors, distinguishing them from other soft tissues in the head. The

proposed approaches seek to improve diagnostic efficiency, reduce reliance on manual

interpretation, and enhance the overall quality of brain tumor analysis in MRI scans.
 LITERATURE SURVEY
Key
Method/ Brain Tumor Accuracy/
Author(s) Technolo
Approach Type(s) Results
gy/Model
Diagnosing
VGG16
Alzheimer's Not specified in
Jain et al. [1] Alzheimer's (pre-
from MRI scans terms of accuracy
trained)
using VGG16
Comparative
study for GoogLeNet
AlexNet,
Yang et al. [2] glioma using Glioma outperforms
GoogLeNet
AlexNet and AlexNet
GoogLeNet
CNN for tumor
Meningioma, CNN (no Training accuracy:
Abiniwanda et identification
Glioma, segmentati 98.51%, Validation
al. [3] without prior
Pituitary on) accuracy: 84.19%
segmentation

SVM,
SVM and PSO
Particle
Zhang et al. for classifier
Not specified Swarm Not specified
[4] design for brain
Optimizatio
MR images
n
 Brain Key Accura
Method/
Author(s) Tumor Technology/Mode cy/
Approach
Type(s) l Results
Particle swarm
optimization with
Not PSO, Artificial Bee Not
Wang et al. [5] artificial bee colony
specified Colony, FNN specified
and feed-forward
neural net

Wavelet entropy
Saritha et al. with PNN Not Not
Wavelet Entropy, PNN
[8] (Probabilistic Neural specified specified
Network)

2-D DWT for feature

extraction and Not DWT, AdaBoost, Not
Nayak et al. [9]
AdaBoost with specified Random Forest specified
Random Forest

Segmentation and
Sheikh
classification with a Not CNN, Stochastic Not
Basheera and
pre-trained CNN specified Gradient Descent specified
Ram [10]
using SGD
 Brain Key Accurac
Method/
Author(s) Tumor Technology/Mod y/
Approach
Type(s) el Results
Particle swarm
optimization with
Wang et al. Not PSO, Artificial Bee Not
artificial bee
[5] specified Colony, FNN specified
colony and feed-
forward neural net
Brain MRI
Khwaldeh et classification into Normal, Not
AlexNet (adapted)
al. [11] normal/abnormal Abnormal specified
and grading
Data augmentation
and pre-trained
Sajjad et al. Multi-grade Not
VGG-19 CNN Model VGG-19 (pre-trained)
[12] cancers specified
for multi-grade
cancer
Pituitary adenomas
tumor Accuracy:
Carlo et al. classification using Pituitary Multinomial Logistic 92%,
[13] multinomial adenomas Regression, KNN AUC:
logistic regression 98.4%
and KNN
Image processing Accuracy:
methodology with Not 94.39%,
Das et al. [14] CNN
CNN for tumor specified Precision:
 PROPOSED SYSTEM

MR
Images Pre-Processing

Tumor
Detection

Tumor
Classification
Using SVM
Classifier Grade of Tumor

Fig1.Proposed System for Brain Tumar Detection

 Data Acquisition:-
The system primarily relies on MRI (Magnetic Resonance Imaging) scans of
the brain, which provide detailed images of the brain structure. MRI scans are
commonly used as they offer better soft tissue contrast, which is crucial for
detecting brain tumors.
 Image Preprocessing:-Normalization and Resizing
The collected images are resized into the input size of the pre-trained network.
The image size required by AlexNet is 227 X 227 X 3, where 3 represent the
number of color channels. The image size required by VGG16 and ResNet50
is 224 X 224 X 3.
The pre-trained networks employ large amounts of data; if this is executed
with small data, this may lead to overfitting of the network which can be
avoided either with dropout layers or by using larger training dataset. In the
proposed work, the data augmentation technique has been chosen to increase
the data volume without muddling the pretrained architecture except the last
layer. Data augmentation method helps in producing a large amount of data
from the small collected data. Augmentation employs operations such as
flipping, rotation, etc. to increase the volume of data.
 Tumor Detection:-

Convolutional Neural Networks (CNNs) are ideal for image

classification tasks. The model can be trained to distinguish between
healthy brain tissue and areas of the brain that may contain tumors.

The initial layers contain low-level features, and the deeper layers
have high-level features. The last fully connected layer of the
network is utilized as a feature vector for extracting the features. In
the modified pre-trained architecture of the proposed framework the
SoftMax classification output layer is replaced with an SVM
classifier for classification purpose. SVM classifier is most widely
used for classification and regression purposes.
 Transfer learning
 Transferring the knowledge, that is already gained in one area to another area

for classification and feature extraction purpose is called transfer learning.

 Transfer learning can be used on the pre-trained models such as AlexNet ,

ResNet and, VGG , etc. A pre-trained network is a deep CNN model which is

trained already over a huge dataset.

 In this method, the final layers of the pre-trained network such as Softmax

activation layer and classification output layers are replaced with SVM

classifier. Instead of training the entire network from scratch, we are using the

already adjusted weights of the network, which is already trained on another

problem. So, computation time is reduced and at the same time performance
 ResNet
 The ResNet secured first position on ImageNet Large Scale
Visual Recognition Challenge (ILSVRC) and Common
Objects in Context (COCO) competition in 2015.
 Due to the increase of layers in a deep network, a
degradation problem occurs. The weights of the layer
cannot be updated correctly to the next layer. To eliminate
this degradation problem, ResNet uses short connections in
parallel to the normal convolutional layers.
AlexNet
 AlexNet is a deep CNN architecture that secured first on the

ILSVRC competition in 2012, trained on 1.2 million images with

1000 different categories using a batch size of 128 which helps to

get stabilized output.

 The modified AlexNet has 5 convolutional layers, 3 max pooling

layers in the original AlexNet are replaced with Average pooling

layers without changing stride value, 3 fully connected layers and

SVM classifier at output layer. RelU activation layer is followed by

every convolutional layer.

VGGNet
VGGNet is a deep CNN architecture. It secured second
place on the ILSVRC competition in 2014. The modified
VGG16 model has 16 convolutional layers, 5 max-
pooling layers, 3 fully connected layers, and the Softmax
classifier is replaced with SVM classifier in the last layer.
The RelU activation layer is followed by all fully
connected layers. The input image size for this network
is 224 X 224 X 3.
SUPPORT VECTOR MACHINES
(SVMS)
 Support vector machines (SVMs) are a set of
supervised learning methods used for classification,
regression and outliers detection.
 The advantages of support vector machines are:

1. Effective in high dimensional spaces.

2. Still effective in cases where number of dimensions is
greater than the number of samples.
3. Uses a subset of training points in the decision
function (called support vectors), so it is also memory
efficient.
SUPPORT VECTOR MACHINES
(SVMS)
 DataSet
Two publicly available MRI datasets are used for validation of the proposed

methodology.

The first brain tumor dataset is collected from Kaggle, and the second brain tumor

dataset is collected from the Multimodal Brain Tumor Image Segmentation Challenge

2015 (BRATS).

Kaggle dataset contains totally 253 MRI images, where 98 of them are non tumor

(normal), and the rest 155 images are Tumor (abnormal).

BRATS 2015 dataset contains 332 MRI brain images, where 156 are LGG (lower grade

glioma), and the rest 176 images are HGG (higher grade glioma).

The samples for the two datasets are exhibited in Figure 2. Each brain tumor dataset is

parted into two sets for training and testing. The training and testing set comprises 70

percent and 30 percent, of the dataset respectively.

Fig2 .Sample of Dataset Used
SOFTWARE AND HARDWARE
SPECIFICATIONS

 Software Specifications:-

 Programming language:- Python

 Liberaries Required:- Open CV, Sci-Kit Learn,Keras

 FrameWork:-Tensorflow

 Editor:- Google Colab/JupitorNote Book(Anaconda)

 Hardware Specifications:-

 System : Pentium i3 Processor,GPU

 Hard Disk : 500 GB.

 Monitor : 15’’ LED.

 Input Devices : Keyboard, Mouse.

 Ram : 8 GB
 APPLICATIONS

 Early Detection of Brain Tumors

 Tumor Classification (Benign vs. Malignant)

 Tumor Segmentation and Localization

 Monitoring Tumor Progression

 Support for Radiologists and Clinicians

 Assist in Personalized Treatment Plans

 Improved Prognosis Prediction

 Support for Research and Clinical Trials

 Assist in Multi-Modal Imaging Integration

 Real-time Diagnostic Assistance in Emergency Settings

 Educational and Training Tool

 ADVANTAGES
 Automated Diagnosis

 High Accuracy and Precision

 Handling Large Datasets

 Reduction of Subjectivity

 Cost-effective

 Real-time Analysis

 Improved Tumor Segmentation

 Continuous Learning and Improvement

 Multimodal Integration
CONCLUSION
In this framework, the Transfer learning process is effectively applied on the

three pre-trained models-AlexNet, ResNet-50, and VGG-16 to classify the brain

tumors. The performance metrics of the above three models are evaluated. The

simulation results demonstrates that ResNet50 performs better in classifying brain

tumor when compared with the other two networks. The proposed model

comprising of Kaggle and BRATS datasets achieved a highest Classification

Accuracy with less computational time after training the framework with data

augmentation and SVM classifier. Therefore, this work can be successfully

employed in the medical field to detect brain tumor. In future, these models can be

used for in-depth classification of the tumor with more subclasses in less

computational time.
REFERENCES
[1] A. Nag et al., "TumorGANet: A Transfer Learning and Generative Adversarial
Network- Based Data Augmentation Model for Brain Tumor Classification," in IEEE
Access, vol. 12, pp. 103060-103081, 2024, doi: 10.1109/ACCESS.2024.3429633
[2] N. Shamshad et al., "Enhancing Brain Tumor Classification by a Comprehensive
Study on Transfer Learning Techniques and Model Efficiency Using MRI Datasets,"
in IEEE Access, vol. 12, pp. 100407-100418, 2024, doi:
10.1109/ACCESS.2024.3430109
[3] Kuraparthi, S., Reddy, M.K., Sujatha, C.N., Valiveti, H., Duggineni, C., Kollati,
M., Kora, P., V, S. (2021). Brain tumor classification of MRI images using deep
convolutional neural network. Traitement du Signal, Vol. 38, No. 4, pp. 1171-1179.
https://doi.org/10.18280/ts.380428
[4] Jain, R., Jain, N., Aggarwal, A., Hemanth, D.J. (2019). Convolutional neural
network based Alzheimer’s disease classification from magnetic resonance brain
images. Cognitive Systems Research, 57: 147-159.
https://doi.org/10.1016/j.cogsys.2018.12.015
[5] Yang, Y., Yan, L.F., Zhang, X., Han, Y., Nan, H.Y., Hu, Y.C., Ge, X.W. (2018).
Glioma grading on conventional MR images: A deep learning study with transfer
learning. Frontiers in Neuroscience, 12: 804. https://doi.org/10.3389/fnins.2018.00804
 [6] Abiwinanda, N., Hanif, M., Hesaputra, S.T., Handayani, A., Mengko, T.R. (2019).
Brain tumor classification using convolutional neural network. In World Congress on
Medical Physics and Biomedical Engineering, 2018: 183-189.
https://doi.org/10.1007/978-981-10-9035-6_33
[7] Zhang, Y., Wang, S., Ji, G., Dong, Z. (2013). An MR brain images classifier system via
particle swarm optimization and kernel support vector machine. The Scientific World
Journal, 2013: 130134. https://doi.org/10.1155/2013/130134
[8] Wang, S., Zhang, Y., Dong, Z., Du, S., Ji, G., Yan, J., Yang, J., Wang, Q., Feng, C.,
Phillips, P. (2015). Feed‐ forward neural network optimized by hybridization of PSO and
ABC for abnormal brain detection. 1178 International Journal of Imaging Systems and
Technology, 25(2): 153-164. https://doi.org/10.1002/ima.22132
[9] Sravan, V., Swaraja, K., Meenakshi, K., Kora, P., Samson, M. (2020). Magnetic
resonance images based brain tumor segmentation-a critical survey. In 2020 4th
International Conference on Trends in Electronics and Informatics (ICOEI)(48184), pp.
1063-1068. https://doi.org/10.1109/ICOEI48184.2020.9143045
[10] Ranjan Nayak, D., Dash, R., Majhi, B. (2017). Stationary wavelet transform and
AdaBoost with SVM based pathological brain detection in MRI scanning. CNS &
Neurological Disorders-Drug Targets (Formerly Current Drug Targets-CNS & Neurological
Disorders), 16(2): 137-149.
[11] Saritha, M., Joseph, K.P., Mathew, A.T. (2013). Classification of MRI brain images
using combined wavelet entropy based spider web plots and probabilistic neural network.
Pattern Recognition Letters, 34(16): 2151-2156.
https://doi.org/10.1016/j.patrec.2013.08.017
 [12] Nayak, D.R., Dash, R., Majhi, B. (2016). Brain MR image classification using two-
dimensional discrete wavelet transform and AdaBoost with random forests. Neurocomputing,
177: 188-197. https://doi.org/10.1016/j.neucom.2015.11.034
[13] Basheera, S., Ram, M.S.S. (2019). Classification of brain tumors using deep features
extracted using CNN. In Journal of Physics: Conference Series, 1172(1): 012016.
https://doi.org/10.1088/1742-6596/1172/1/012016
[14] Khawaldeh, S., Pervaiz, U., Rafiq, A., Alkhawaldeh, R.S. (2018). Noninvasive grading of
glioma tumor using magnetic resonance imaging with convolutional neural networks. Applied
Sciences, 8(1): 27. https://doi.org/10.3390/app8010027
[15] Sajjad, M., Khan, S., Muhammad, K., Wu, W., Ullah, A., Baik, S.W. (2019). Multi-grade
brain tumor classification using deep CNN with extensive data augmentation. Journal of
Computational Science, 30: 174-182. https://doi.org/10.1016/j.jocs.2018.12.003
[16] Carlo, R., Renato, C., Giuseppe, C., Lorenzo, U., Giovanni, I., Domenico, S., Mario, C.
(2019). Distinguishing functional from non-functional pituitary macroadenomas with a machine
learning analysis. In Mediterranean Conference on Medical and Biological Engineering and
Computing, Springer, 76: 1822-1829. https://doi.org/10.1007/978-3-030-31635-8_221
[17] Das, S., Aranya, O.R.R., Labiba, N.N. (2019). Brain tumor classification using
convolutional neural network. In 2019 1st International Conference on Advances in Science,
Engineering and Robotics Technology (ICASERT), pp. 1-5.
https://doi.org/10.1109/ICASERT.2019.8934603
[18] https://www.hopkinsmedicine.org/health/conditions-and-diseases/brain-tumor/brain-tumor-
types