Automating and Improving Brain Tumor Classification Using Novel Convolutional Neural Network Model

Abstract

With each second that passes patients who suffer from brain cancer become more at risk. Delays and misinterpretations from radiologists of brain tumor MRIs (magnetic resonance images) will affect the administration of treatment, endangering patients. It is necessary to consider methods that minimize fallacies present while maximizing time efficiency. This project aims to address the problem using AI classification models, in particular, convolutional neural networks (CNNs) for the purpose of distinguishing between tumor types. This project successfully develops a novel model that is capable of distinguishing—using MRI scans— between glioma, meningioma, pituitary tumors, and unaffected brains. Analysis of the model concluded that it was accurate and consistent: 91 percent accurate and F1-score of 0.92. This displays that CNNs can be used for brain tumor classification and other medical environments for diagnosis. At minimum, they can be used as an aid to experts in the field for more conclusive diagnosis.