Validating Explainable AI in Clinical Neuroimaging

01/01/2026 to 31/12/2028

While deep learning methods achieve high predictive performance in neuroimaging, their clinical and scientific use is limited by the lack of validated approaches for model interpretation. This project develops a rigorous, domain‑specific validation framework for explainable AI in neuroimaging by generating anatomically localized ground‑truth targets from imaging‑derived phenotypes and systematically benchmarking explanation methods across architectures and clinically relevant patterns. The project will deliver validated benchmarks, open tools, and best‑practice guidelines to enable reliable and biologically meaningful interpretation of deep learning models in neuroscience and clinical research.