Brain Tumor Detection with Explainable AI: A GRAD-CAM Approach to Medical Imaging

The early identification and management of brain tumours require accurate and interpretable MRI images. The study used a VGG16-based CNN to categorise brain tumours into four categories and Gradient weighted Class Activation Mapping (Grad-CAM) to enhance explanations. Most of the deep learning models are generally considered as black box models, making decision making difficult to understand. Grad-CAM focus on key areas in MRI images that mostly impact the predictions, hence improving transparency and clinical interpretation. Preprocessing approaches such as data augmentation and data normalization enhance the generalization of the model. Class balancing methods are used to sort out data imbalance and avoid biased predictions. Here, early stopping and model checkpointing are also implemented to avoid overfitting, and the model is trained with mixed precision for optimal computational efficiency. Both training and test datasets are used to assess the model performance on various images. The results shows excellent classification accuracy of 99.1% on test dataset. The trust in deep learning based models is increased through the combination of high accuracy and interpretable AI approach in this study. Such an approach improves its applicability in real health care settings as it closes the gap between AI predictions and clinical decision making processes.