Brain Tumor Classification Using Explainable Artificial Intelligence

This work discusses a methodology for brain tumor classification using a Convolutional Neural Network (CNN) integrated with explainable AI techniques to enhance model transparency. Using a Kaggle magnetic resonance (MRI) dataset with two classes, tumorous and non-tumorous, this study addresses the challenge of class imbalance through extensive data augmentation, increasing the dataset to approximately 1,000 images per class. After preprocessing, a custom CNN model was trained, achieving a classification accuracy of 90.88%, demonstrating its effectiveness in distinguishing between tumor classes. The model architecture included convolutional, pooling, dense and dropout layers optimized for robust performance on medical imaging data. SHapley Additive Explanations (SHAP) is applied to explain the result of the model, generating visual insight into the features that affect the classification results. This interpretability is critical for clinical applications, allowing healthcare providers to understand and trust AI-assisted decisions in diagnostics. Although the study shows promising results, further validation with extensive and heterogeneous datasets is suggested. The integration of CNNs and XAI techniques like SHAP offers a powerful and transparent framework for brain tumor detection, aiming to support clinicians with reliable and interpretable AIbased diagnostic tools.