Interpretable deep learning for enhanced clinical decision making in brain tumour detection using explainable AI

Deep learning has significantly improved the detection of brain tumors, but its “black box” nature essentially limits its application Nevertheless, the new approach indicates an open and clinician-centric perspective, which is a direct result of the integration of explainable AI (XAI) with a quantum-inspired optimization technique to resolve this issue. It is the Quantum Dragonfly Algorithm (QDA) that is at the center of the work which is the closest simulation to dragonfly swarming behavior for feature selection optimization and thus tumor segmentation from MRI is improved. QDA has the capability to explore the entire space of solutions without falling into the trap of poor local optima and therefore, reach the highest possible accuracy. To be granted trust by the clinic and also to be understood, the system provides Grad-CAM and SHAP explainability. Grad-CAM is a technique that highlights the most important areas for a given classification on the MRI while SHAP provides complete and detailed feature attributions for the model's decisions which, in result, are transparent and interpretable. Besides that, an Interactive Visualization Dashboard takes even more steps to bridge the gap by giving the clinicians access to the results, explanations, and their feedback which is an indication of their continuous learning and adjustment right there. The method can be considered as the combination of a dependable, clear, and user-friendly tool for brain tumor detection that incorporates the qualities of high performance and clinical trust.