Automated Brain MRI Tumor Classification and Visual Explainability Using Transfer Learning with Xception CNN and Grad-CAM

Spotting a brain tumor early gives doctors a crucial head start when deciding how to treat a patient. In everyday hospital work, Magnetic Resonance Imaging (MRI) is the preferred scan for looking at brain anatomy and flagging unusual growths. Even so, reading through stacks of MRI slices by eye takes hours of focused effort, and two doctors looking at a borderline scan might not always reach the exact same call. To address these challenges, a deep learning-based Brain Tumor Detection and Classification System has been developed using a Convolutional Neural Network (CNN) built on the Xception architecture. Rather than training a blank network from zero, the system applies transfer learning to extract features from brain MRI scans and sort them into four categories: glioma, meningioma, pituitary tumor, and no tumor. Every image goes through resizing and intensity normalization during preprocessing, while brightness data augmentation improves how well the model generalizes to unseen scans. Alongside classification, Gradient-weighted Class Activation Mapping (Grad-CAM) is applied to produce visual heatmaps that highlight the exact scan regions influencing each prediction. The trained model is integrated into a user-friendly web interface where users can upload MRI scans and view the predicted class, class-wise confidence scores, and Grad-CAM heatmaps. Overall, the system is designed to support automated MRI analysis as a computer-aided research tool rather than a replacement for professional medical diagnosis.