Task 2: MRI + Biopsy Risk Stratification¶

Goal: After biopsy, integrate MRI findings, biopsy pathology whole-slide images, and PSA-related variables to recommend a management strategy, with guideline-based reasoning (EAU/NCCN).

Input data

• structured-prompt.json: structured clinical variables, such as PSA, PSA density, PI-RADS category, biopsy ISUP grade group, primary and secondary Gleason patterns, comorbidities, and csPCa likelihood score
• prostate-treatment-decision-clinical-data.json: radiology report, biopsy pathology report, prior clinical notes, PSA trend, laboratory results, and family history
• prostate-modality-level-neural-representations.json: MRI and biopsy-slide neural representations

See Data Sources and Imaging for the complete input structure and variable descriptions.

Output required

• Treatment recommendation, one of four categories: active_surveillance, active_treatment, continued_surveillance, watchful_waiting
• Structured reasoning including confidence in the treatment recommendation, the importance assigned to the relevant clinical variables, the clinical-data sections/tools accessed by the agent during the decision, and a free-text explanation of the main factors supporting or contradicting the chosen strategy.

Ground truth: Derived retrospectively from biopsy histopathology and PSA-related clinical variables per EAU/NCCN guidelines.

Expected output per case: Two JSON files (~5 KB) containing the treatment recommendation and structured reasoning trace.

Provided neural representations: MRI features from an MRI-based prostate cancer detection model and biopsy WSI features from an automated Gleason grading model.

Primary metric: Case score, please refer to this page and the repo.