Explainable AI and deep learning for brain tumour classification using grad-CAM visualization

Brain tumours rank as the fifth most common cancer type, affecting approximately 90,000 people annually and pose high risks due to their potential for life-threatening neurological impacts. Traditional diagnostic methods often lack visualization clarity, impeding accurate interpretation and delaying critical treatment decisions. To address these challenges, this study introduces a robust deep learning framework that combines ResNet101, InceptionV3, DenseNet201 and VGG16 architectures to classify brain tumours into four categories: pituitary, meningioma, glioma and no tumour. The model was trained and validated on a dataset of 7,022 MRI images and was finely tuned to maximize classification accuracy. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to generate interpretable heatmaps, highlighting key regions in MRI scans that influence classification. This enhanced interpretability facilitates clinical trust and understanding by allowing the model's focus areas to align with clinically relevant regions. The combined model achieved a highly competitive performance, with 99.2% accuracy, 98.8% precision, 98.7% recall and an F1-score of 99.1%, along with an AUC of 1.00 across all classes, reflecting robust diagnostic potential. Grad-CAM was specifically applied to tumour classes, revealing critical areas in MRI scans and supporting improved interpretability. This framework marks a substantial advancement in brain tumour classification, streamlining the diagnostic process, minimizing reliance on manual interpretation and fostering more efficient, accurate and sustainable healthcare solutions.