Explainable CNN for Brain Tumor Detection and Classification
Abstract
Image analysis is an important part of medical decision-making, where it helps clinicians to reach a diagnosis and plan treatment. Artificial intelligence (AI) and machine learning (ML) methods for medical image analysis have grown rapidly. Explainable artificial intelligence (XAI) methods aim to make AI applications more usable and trustworthy by showing how a model reaches its decisions. This paper reviews recent XAI methods and their use in medical image analysis, including rule-based systems, attention mechanisms, and feature visualization. A convolutional neural network (CNN) was used to classify brain magnetic resonance imaging (MRI) scans by tumor type, and an XAI model explains how the network uses the images to reach each prediction. Of the 15,030 images, 13,563 were used for training and 1,467 for validation and testing. The model reached an accuracy of 80.55%. Local Interpretable Model-Agnostic Explanations (LIME) is used as the example XAI model.