Following our recent article on the risks of shortcut learning in AI-driven cancer diagnostics, join us for this short Pathology News interview, where Sidney Ocanagil-Tunstall speaks with Dr. Fayyaz Minhas from the Tissue Image Analytics Centre (University of Warwick).

In this brief conversation, Dr. Minhas expands on the key themes discussed in the article, exploring how AI models in pathology can achieve strong predictive performance while relying on correlations rather than true biological signals. He discusses challenges such as shortcut learning and confounding factors, and what this means for applications like predicting mutation status from whole slide images.

The interview also touches on the importance of robust validation, thoughtful data design, and stronger collaboration between clinicians and computational scientists to ensure AI tools are safe, reliable, and truly beneficial for patient care.