Brain Tumor Detection from MRI Using 2D Convolutional Neural Network

Date
2021
Research Supervisor
Editor
Journal Title
Volume
Issue
Journal Title
Journal ISSN
Volume Title
Abstract
Brain Tumor are considered to be one of the most aggressive forms of tumors. But early detection of brain tumors can raise patient survival expectations significantly. That’s why accurate and quick detection is a must for patient survival. The traditional way of Brain Tumor detection is doing an MRI of the brain. These MRI images reveal the abnormal cells, which suggest whether the patient is suffering from Brain Tumor or not. But detecting Brain Tumor in the traditional way is very costly, time-consuming, and complex. That’s why we propose a Deep Learning based framework to automatically detect Brain Tumor. Deep Learning has already proven to be effective in diagnosing medical images. In this project, we used 2D Convolutional Neural Network (CNN) for the detection. It is a Deep Learning technique for image classification. For training and testing purpose, our proposed 2D CNN architecture used an online dataset named “Br35H :: Brain Tumor Detection 2020”, which includes 1500 positive cases and 1500 negative cases of Brain Tumors MRI. These 3000 MRI images will be divided into 2 parts. 70% for the training set (2100 MRIs), 30% for the test set (900 MRIs). Here, we made sure train and test dataset are balanced, that means positive and negative case ratio is maintained. Then we preprocessed the dataset from MRI to tensor, so that we could feed the them as input. Upon preprocessing the images, we used the training portion to train some CNN models. We also train our dataset on some popular deep learning pre-trained models (AlexNet, VGG, ResNet, DenseNet). Then selected the best model based on the training score. Finally, we used the test set for the evaluation. As it is a binary classification, we use the accuracy and F-1 Score for the selection of the CNN model. With our ResNet50 pretrained model, we could able to achieve an accuracy score of 97.22% and F-1 score of 97.22% on the test set.
Description
Keywords
Citation
Department Name
Electrical and Computer Engineering
Publisher
North South University
Printed Thesis
DOI
ISSN
ISBN