Brain Tumor Detection On Magnetic Resonance Imaging Using Ai

Abstract

Early detection and classification of brain tumor grade are crucial for improving treatment and potentially prolonging survival rate. This work aims to address this issue using MRI-based classification and to create an accurate model for the diagnosis of gliomas. This research presents a deep learning pipeline consisting of three main steps: (1) MRI brain tumor images are segmented using preprocessing methods with two custom UNet architecture; (2) Collect regions of brain tumor are extracted bases on the prediction of segmentation; and first approach (3) HGG and LGG are classified through the implementation of a custom CNN model architecture; second approach (4) Grade 1, grade 2, grade 3 are classified by using pre-trained VGG-19. The proposed network structure achieves a significant performance, with the best overall accuracy of 98.9 % and 85.14 %, respectively, for 2 classes and 3 classes. This model can potentially be integrated into a computer-aided diagnosis system that supports clinicians in the diagnosis of gliomas.