Investigating Deep Learning Models for the Classification of Brain Tumors to Infer Biological Significance from MRI Images

Brain tumors can pose a dangerous threat if misclassified. The primary way of classifying these brain tumors is through Magnetic Resonance Imaging (MRI) images, which is a highly subjective and manual process. With improved technology and machine learning, this paper proposes CapsNet model and ViT model, which differs from traditional CNN model, that aims to perform better on small image size, long-range dependencies and sequential features, which human tissue often contains. After extensive training and evaluation, the results of the CapsNet models and ViT models on the testing dataset come out to be 74.81% and 93.84%, respectively. We also find that the potential of our work will be enormous in propelling the actual application of ML in clinical diagnosis because the incredible accuracy will protect patients and doctors from the risk of misdiagnosis. Furthermore, interpreting the models also help us reinforce and gain new perspectives on clinical diagnostics.