Abstract:
In this study, we propose an automated approach for brain tumor segmentation and
classification from MRI images which is a leading sector in medical imaging research
because of the importance they have for patient diagnosis and therapy. This project does a
comprehensive literature review and provides approaches to improve the accuracy and
efficiency of brain tumor detection. The dataset, sourced from Kaggle, consists of 4237
MRI images categorized into four classes: No Tumor, Glioma, Meningioma, and Pituitary
tumors. Corresponding segmentation masks are also included for precise delineation of
tumor regions. The detection system's performance is improved through the use of
techniques such as feature extraction, image resizing, and grayscale conversion. Several
segmentation and classification methods, including Convolutional Neural Networks
(CNNs), Artificial Neural Network (ANN), Graph Neural Network (GNN) are evaluated.
The segmentation task is performed using a customized CNN, achieving an impressive
accuracy of 96.32%. For the classification task, a variety of machine learning models were
tested, including Random Forest, Extra Trees, Logistic Regression, Gradient Boosting,
Support Vector Machine, Decision Tree, Naive Bayes, ANN, and GNN. Among these, the
Graph Neural Network (GNN) demonstrated superior performance, achieving an accuracy
of 99.65%. The project also involves developing a web application where users can upload
MRI images to receive automatic tumor detection and classification. This application aims
to assist medical professionals by providing quick and accurate diagnoses, potentially
improving patient outcomes. The project aims to overcome existing obstacles in brain
tumor identification while also contributing to the development of accurate and trustworthy
diagnostic tools for healthcare professionals.