BRAIN TUMOR

DETECTION USING
MACHINE LEARNING
Prepared By: Muhammad Saad Aziz
Ulster University
Table of Contents

1.Introduction
2.Literature Review
3.Problem statement
4.Aims and objective
5.Methodology
6.Summary
2
 Introduction

• Preliminary findings from a nationwide study show differences

in the types and ages of brain tumor patients in Pakistan
compared to the developed world, highlighting the need for
specialized care.(Agha Khan)
• A brain tumor is a serious disease, and if the necessary
intervention is not done, it can eliminate functions such as
walking and eating that meet the person's basic needs.

4
 GLIOM
A
Gliomas are a type of brain tumor that
originates from glial cells, which support
and protect neurons in the central
nervous system. They can be benign or
malignant and are among the most
common types of brain tumors
Meningio
Meningiomas are typically slow-growing brain tumors that
ma
form in the meninges, the protective membranes
surrounding the brain and spinal cord. Most meningiomas
are benign and may not cause symptoms unless they grow
large.

Pituitar
Pituitary tumors develop in the pituitary gland, a
y
small gland located at the base of the brain. These
tumors can disrupt hormone regulation and have a
range of effects on the body.

6
 Literature Review
Since the 1920s, epidemiological studies on open-angle brain tumors
have significantly contributed to brain tumor research. Early studies
often conflated brain tumors with ocular hypertension and lacked
clear disease criteria.

Recent research has applied machine learning, such as convolutional

neural networks (CNN) and support vector machines (SVM), for brain
tumor detection and classification, achieving notable success rates.
Various studies have utilized advanced image processing and
segmentation techniques, demonstrating their effectiveness in MRI-
based brain tumor classification.

7
 Literature Review
MRI-based brain
The developed networks achieved optimal accuracy of approximately
tumor detection
80% to 86%, and areas under the receiver operating characteristics
Badža and Barjaktarovic´ using convolutional
curves (AUROC) are 0.88 or 1. Performance analysis proposes a
deep learning
renovation of our proposed techniques by comparing related papers.

The Watershed procedure is employed to accurately segment the

CNN identifying tumor area, with CNN identifying tumor and non-tumor regions in brain
tumor and non- MRI images. Additionally, a dynamical direction estimation model
Watershed
tumor regions in extracts structural features of the brain. Successful testing is
brain MRI images. conducted on previous Brats datasets, achieving a classification
success rate of 89.03%.

convolutional
neural networks
convolutional neural networks (CNN) and vector machines (VM) to
(CNN) and vector
classify brain tumor images into three subtypes: meningiomas, and
Nabeel Salih Ali Zahraa M machines (VM) to
gliomas glands. The model is tested on fifty MRI images and achieves
classify brain tumor
an overall accuracy of 80%.
images into two
subtypes

8
 Study Dataset Techniques Performance

Multi-level thresholding
Younis, A., Qiang, L., BRATS 2015 JI 81.94%, DSC
with level-set
Nyatega 2022 BRATS-MICCAI 89.91%
segmentation

Dehkordi, A. A., Hashemi

https://radiopaedia.org K-means and FCM ACC 56.4 %
2022

Patch based k-means with

T. A. Jemimma BRATS SI 91%
FCM

. Dandıl, E., & Karaca BRATS 2012 Random DSC 91%

Amin, Javaria & Sharif,

BRATS Force Clustering -
Muhammad

Mawaddah, S., Mufid BRATS 2017 Random DSC 62.5%

9
 Literature Review
Summary
•The proposed model was tested only on a small dataset, and its accuracy on
larger datasets remains untested. Additionally, the model requires high
computational power, limiting its practical application.
•The model requires manual feature extraction, which can be time-consuming
and subjective. The study also lacks comparative analysis with other models.
•The dataset used in the study is relatively small, limiting the generalizability of
the model. The proposed model also requires further validation on a larger
dataset.
•The study does not provide detailed information on the choice of
hyperparameters, which can have a significant impact on the model's
performance. Additionally, the dataset used in the study is relatively small,
limiting the model's generalizability.
10
Problem statement

The problem involves brain tumors impeding vital brain

functions, leading to complications and mortality.
Insufficient patient data hinders prognosis accuracy. The
solution aims to create an automated system for continuous
brain tumor monitoring, focusing on unbalanced image
datasets. There's currently no deep learning-based
framework for effective tumor identification. The challenge
includes the lack of scalable healthcare systems and the
need for a web and mobile app to store patient data,
prioritize treatment, and provide timely alerts for remote
monitoring.

11
 AIM

This project aims to develop an expert system using

Convolutional Neural Networks for precise diagnosis and
classification of brain tumor diseases, complemented by a
scalable mobile app that leverages patient clinical data to
enhance treatment prioritization and healthcare resource
utilization, ultimately improving patient outcomes.

12
 Objectives

1 Create a novel CNN architecture optimized for brain

tumor classification and early detection.

Develop a mobile app that can analyze brain tumor

images and provide diagnostic results for enhanced
2
accessibility to healthcare services.
METHODOLOGY

14
 Methodology
1.Data Loading and Exploration: Load the brain MRI dataset, visualize
sample images, and gain a better understanding of the data
distribution.

15
 Methodology
Data Preprocessing: Prepare the data for training by performing
image resizing, normalization, and augmentation techniques.

16
 Methodology
Model Architecture: Design a CNN architecture suitable for brain tum
classification and implement it using a deep learning framework.

17
 Methodology
Model Training: Train the CNN model on the preprocessed training
data and monitor its performance on validation data.

18
 Methodology

• Model Evaluation: Evaluate the trained model on the test set to

assess its performance and generalization capabilities.

• Mobile App Integration: Develop a user-friendly mobile app that

allows users to upload brain MRI images. Implement a feature to
process the uploaded images using the trained CNN model and display
diagnostic results.

19
• Incorporation of batch normalization for improved performance.
• Implementation of max pooling and ReLU activation functions in the down-
sampling process.
• Structuring the architecture with a series of convolutional layers organized in
blocks to capture recurring patterns.
• Notable decrease in validation accuracy for brain tumor detection and
classification when reducing a block to just one convolution layer.

20
21
 Conclusion

A mobile application has been developed, enabling users

to upload brain images for tumor detection, with results
promptly delivered. This innovation represents a
significant advancement in the field, offering a user-
friendly and efficient means of brain tumor diagnosis. The
thesis findings highlight the effectiveness of Convolutional
Neural Networks, achieving remarkable accuracy rates,
further emphasizing the potential impact of this work in
improving disease detection and patient care.

24
 References
1.Denis Schapiro et al. ”a modular, scalable pipeline for image processing for
multiplexed tissue imaging” featured in methods in cell biology in 2022.
2.Boran Sekeroglu Ahmet Ilhan and Rahib Abaiyev. ”Nonpara?metric
Localization and Enhancement Methods with U-Net for Brain Tumor
Segmentation in MRI Images”. published in the International Journal of
Computer Assisted Radiology and Surgery in 2022.
3.Dhimas Arief Dharmawan. ”evaluating the fairness of publicly available
retinal blood vessel segmentation algorithms in terms of performance
evaluation”. 2021
4.Chengxiang Qiang Li and Ching-Hsin Wang. ”Automated 3D Multi-Channel
Feature Selection Convolutional Neural Network for Brain Tumor
Segmentation” which appeared in the Journal of Imaging Science and
Technology in 2022
 Thank you
For your attention

26