Evidential Quantification of Uncertainty for Reliability in XAI Saliency Maps – EQUR-XSM

EQUR-XSM (Evidential Quantification of Uncertainty for Reliability in XAI Saliency Maps) is a research project focused on developing trustworthy, interpretable AI solutions for medical image analysis. By combining Evidential Deep Learning (EDL) with advanced uncertainty modeling, the project aims to quantify and calibrate model confidence, reducing the risk of overconfident misdiagnoses. EQUR-XSM also explores explainability via saliency maps enhanced with pixel-wise uncertainty, offering clinicians clear insights into which areas of an image are both critical and trustworthy. Further, the project targets multi-modal fusion, explicitly modeling modality-specific uncertainties (e.g., MRI vs. CT) to achieve robust segmentation for complex cases like tumors, skeletal muscle abnormalities, and hematological diseases. Collaborations with clinical partners (CHU Grenoble and Nantes) will support real-world validation, while open-source data will ensure broad reproducibility. Through this approach, EQUR-XSM seeks to bridge methodological gaps in calibration, XAI, and multi-modal integration, ultimately improving patient care.