🦉OWL: Observing What Models Learn¶

Observing What Models Learn (OWL) is a competition focused on evaluating interpretable AI methods for scientific discovery in computational pathology. While modern AI models can achieve strong predictive performance in medical imaging, the explainability of state-of-the-art prognostic AI in histopathology remains largely limited to attention heatmaps. There is no principled framework for assessing whether AI explanations reveal meaningful biological insights or support the discovery of novel disease biomarkers.

OWL addresses this challenge using biochemical recurrence (BCR) prediction from prostatectomy whole-slide histopathology images as a representative scientific discovery task. The competition data consists of patient cases: H\&E-stained prostatectomy slides with associated recurrence status and time to last follow-up. Participants will develop AI models that predict recurrence risk while generating interpretable outputs, including spatial attribution maps and human-readable explanations.