Fusion-Based Brain Tumor Classification Using Deep Learning and Explainable AI, and Rule-Based Reasoning

The ensemble approach significantly outperformed individual models, achieving 91.7% accuracy by combining MobileNetV2 and DenseNet121 through soft voting, demonstrating that combining multiple AI systems can produce more reliable results than relying on a single model for brain tumor classification.

Explainability tools like Grad-CAM++ showed strong alignment with expert knowledge, with visualizations achieving Dice coefficients up to 0.88 and IoU scores up to 0.78 when compared to tumor regions identified by radiologists, suggesting the AI's decision-making process can be meaningfully validated against human expertise.

Radiologists rated the AI explanations as highly useful and trustworthy, with board-certified radiologists giving explanation usefulness an average score of 4.4 out of 5 and heatmap-region correspondence a 4.0 out of 5, indicating that transparent AI systems can build clinical confidence rather than operating as "black boxes."

The framework integrated symbolic clinical rules alongside deep learning, mapping AI predictions to established radiological heuristics, which demonstrates a practical approach to combining AI automation with domain expertise rather than replacing human judgment entirely.

The study used rigorous validation methods including stratified 5-fold cross-validation and evaluation across multiple performance metrics (accuracy, precision, recall, F1-score), establishing a methodologically sound foundation for considering this technology in clinical settings.