Counterfactual-Driven Benchmarking for Explainable AI in Rare Disease Imaging

Both Metsch & Hauschild (2025, BenchXAI) and Jin et al. (2022) point out that existing XAI benchmarks focus on prevalent pathologies and standard datasets, often neglecting the unique challenges of rare disease imaging. Inspired by Shmueli et al. (2025) and Baron (2023), this research would create synthetic or semi-synthetic counterfactuals (e.g., “What would this scan look like if the rare condition were absent?”) and test if current XAI methods can provide coherent, clinically useful explanations in these settings. This could expose fundamental weaknesses in current XAI for rare cases, drive the development of more robust methods, and ultimately improve care for underrepresented patient populations.

References:

1. Evaluating Explainable AI on a Multi-Modal Medical Imaging Task: Can Existing Algorithms Fulfill Clinical Requirements?. Weina Jin, Xiaoxiao Li, Ghassan Hamarneh (2022). AAAI Conference on Artificial Intelligence.
2. BenchXAI: Comprehensive Benchmarking of Post-hoc Explainable AI Methods on Multi-Modal Biomedical Data. Jacqueline Michelle Metsch, Anne-Christin Hauschild (2025). bioRxiv.
3. Evaluating Explainable AI on a Multi-Modal Medical Imaging Task: Can Existing Algorithms Fulfill Clinical Requirements?. Weina Jin, Xiaoxiao Li, Ghassan Hamarneh (2022). AAAI Conference on Artificial Intelligence.
4. From What Ifs to Insights: Counterfactuals in Causal Inference vs. Explainable AI. Galit Shmueli, David Martens, Jaewon Yoo, Travis Greene (2025). arXiv.org.
5. Explainable AI and Causal Understanding: Counterfactual Approaches Considered. S. Baron (2023). Minds and Machines.