Brain MRI Classification and Segmentation of Glioma, Pituitary and Meningioma Tumors Using Deep Learning Approaches

Abstract

This paper explores the application of deep learning approaches to segment and classify brain tumors in MRI images, specifically targeting glioma, meningioma, a nd pituitary tumors. The paper uses the Crystal Clean Brain MRI Dataset, which includes approximately 21,000 pictures of both normal brains and tumor brains. This dataset has been useful in training advanced deep learning architectures such as U-Net, SegNet, ResNet50, and MobileNet. To improve mask generation and tumour segmentation accuracy, our methodology integrates advanced preprocessing techniques such as data augmentation and the novel Segment Anything Model (SAM). These models were then evaluated using a systematic process that included gathering data, preprocessing, mask creation, segmentation, and classification, with the ResNet50 model obtaining t he highest accuracy of 98%. The results of this study greatly improve the accuracy of brain tumor detection and classification, demonstrating the promise for future advances in automated medical imaging.