Exploring Explainable AI in Brain Tumor Detection: A Hybrid Approach Using EfficientNet and CBAM

This research presents an advanced methodology for brain tumour classification using magnetic resonance imaging (MRI), combining the EfficientNet-B0 architecture with Convolutional Block Attention Modules (CBAM). The proposed model effectively captures both channel and spatial attention by replacing the Squeeze-and-Excitation (SE) modules with CBAM, enhancing feature extraction and classification accuracy. The framework is evaluated on a comprehensive dataset comprising four classes: pituitary no tumour, meningioma, and glioma. Leveraging transfer learning and Grad-CAM-based explainability, the proposed model attains an outstanding accuracy of 99%, with weighted F1 score and macro scores exceeding 98%. Comparative analyses confirm its superior performance against state-of-the-art methods, particularly in classifying complex tumour types. The combination of accuracy, efficiency, and interpretability highlights the potential of this approach for real-world medical diagnostics.