Identification and Prediction of Brain Tumor Using VGG-16 Empowered with Explainable Artificial Intelligence

Abstract

The abnormal development of cells is what causes brain tumors. It is one of the world's leading causes of mortality among adults. Brain tumor detection in a timely manner can prevent millions of deaths. Earlier detection of brain tumors using Magnetic Resonance Imaging (MRI) may increase the patient's chance of survival. MRI is the most prevalent diagnostic technique for brain tumors. The enhanced visibility of tumors on MRI facilitates subsequent treatment. Identification and prediction of brain tumors are essential to their diagnosis and treatment. This article presents a study that utilizes the VGG16 deep learning model to classify brain MRI images obtained from a dataset sourced from Kaggle, comprising two classes: normal and tumor. The dataset is separated into training and testing sets, and the VGG16 model is trained to achieve a testing accuracy of 97.33%. Despite the high accuracy achieved, deep learning models like VGG16 are often perceived as "black boxes," providing predictions without clear explanations. To address this limitation, Layer-wise Relevance Propagation (LRP) is applied to the VGG16 predictions to shed light on the decision-making process and provide interpretability.