🧪 Task Description¶

Participants receive Hematoxylin and eosin (H&E) stained pathology whole slide images (WSIs) (see the data release for resolution, file type, and cohort details). For each slide, they build models that generate:

  • Pathology report — automatically generated, structured like real practice and aligned with expert references where provided.
  • Reasoning — pathologist-style explicit diagnostic reasoning (e.g. structured chain-of-thought); format and schema per evaluation package.

In short: from each WSI, the system must produce a pathology report plus explicit reasoning that reflects how a pathologist would justify the conclusions behind that report.


Challenge objectives¶

The challenge is organized around two goals:

  1. Evaluate diagnostic reasoning — measure and compare algorithms on diagnostic reasoning, not only surface report match to a reference.
  2. Generalize across settings — Encourage medical AI models that remain useful across diverse organ specimens and multiethnic populations, not only narrow case types.