Goal: After biopsy, integrate MRI findings, biopsy pathology whole-slide images, and PSA-related variables to determine whether the patient is eligible for active surveillance or should proceed to radical prostatectomy, with guideline-based reasoning (EAU/NCCN).

Input modalities

  • Multiparametric MRI
  • Biopsy H&E whole-slide images (.tif)
  • Clinical variables: PSA, age, Gleason Grade Group, biopsy burden

Output required

  • Binary recommendation: eligible / not eligible for active surveillance
  • Structured reasoning identifying dominant evidence and explicitly flagging factors that contradict active surveillance

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

Dataset

Split Cases Notes
Training 75 Radboudumc; real-world class distribution
Validation 75 Radboudumc; up to 5 submissions allowed
Test 300 100 cases from Karolinska Institute (external); 1 submission allowed

Input data (precomputed): Precomputed MRI features as .h5 files + precomputed biopsy WSI features as .h5 files + clinical variables as .csv. Raw images are not provided for inference.

Expected output per case: JSON file (~5 KB) containing the active surveillance eligibility decision and structured reasoning trace.

Primary metric: AUROC

Provided tools: Automated Gleason grading tool for biopsy WSI; MRI-based prostate cancer detection tool