Goal: After radical prostatectomy, integrate whole-slide images from the prostatectomy specimen, biopsy findings, preoperative MRI, and longitudinal PSA kinetics to estimate time-dependent biochemical recurrence (BCR) risk. Cases may include multiple MRIs and/or biopsies over time, and missing modalities are intentionally present.
Input modalities
- Report of prostatectomy H&E whole-slide images
- Report of preoperative biopsy WSI
- Report of multiparametric MRI
- Longitudinal PSA measurements (prior to surgery)
- Clinical variables: age, Gleason Grade Group, surgical margins, extracapsular extension, pT stage, and T stage
Output required
- Quantitative BCR risk estimates at 1, 2, and 5 years
- Structured prognostic reasoning identifying dominant factors contributing to recurrence risk and handling of missing or conflicting data
Ground truth: BCR defined as confirmed PSA rise >= 0.2 ng/mL post-prostatectomy. Patients without recurrence at last follow-up, as well as those who received salvage treatment, migrated, died, or were otherwise lost to follow-up, are treated as censored observations.
Dataset
| Split | Cases | Notes |
|---|---|---|
| Training | 75 | Radboudumc; real-world class distribution |
| Validation | 75 | Radboudumc; up to 5 submissions allowed |
| Test | 250 | 100 cases from Karolinska Institute (external); 1 submission allowed |
Input data: JSON file with structured input modalities.
Expected output per case: JSON file (~5 KB) containing BCR risk estimates at 1, 2, and 5 years, and a structured reasoning trace.
Primary metric: Time-dependent AUC
Provided tools: Standardized preprocessing and feature extraction for prostatectomy WSI; automated Gleason grading for biopsy and prostatectomy specimens; MRI-based cancer detection tool