What participants receive¶

For all tasks, participants receive a structured JSON file per case containing the available input modalities for that patient. These include clinical variables (e.g. age, PSA, PI-RADS score, DRE results), tool outputs (e.g. deep learning-generated csPCa probability, automated Gleason grading results), and report-level summaries of imaging and pathology findings where applicable. Participants do not receive raw imaging data at inference time; the agent reasons over the structured inputs and tool outputs provided in the JSON.

A full list of available variables and tool outputs is described on the Tools page.

Underlying imaging and acquisition¶

The structured inputs provided to participants are derived from clinical imaging acquired as part of routine prostate cancer care. The table below describes the original imaging devices and protocols for reference.

Data origins¶

CHIMERA-agent builds on the CHIMERA challenge at MICCAI 2025, extending it with agent-based reasoning and sequential clinical decision-making. Data for CHIMERA-agent is collected independently.

All training and validation data originate from Radboud University Medical Center (Radboudumc), Nijmegen, The Netherlands. The test set includes cases from Radboudumc, Canisius Wilhelmina Hospital (CWZ), Karolinska Institute (Sweden), and additional contributing institutes. MRI examinations were interpreted by board-certified radiologists following PI-RADS v2.1. Biopsy procedures were performed by experienced urologists, and histopathology was assessed by board-certified pathologists specializing in genitourinary pathology.

Public datasets recommended for pre-training¶

Participants who wish to train or fine-tune their own modality-specific models (instead of using the provided tools) may use the following publicly available benchmarks:

• PI-CAI: prostate cancer detection in MRI (1,000+ cases)
• PANDA: prostate cancer Gleason grading in WSI (1,000+ cases)
• LEOPARD: biochemical recurrence prediction from WSI (500+ cases)

External data and pretrained models are permitted provided they are freely and publicly accessible under a permissive open-source license (e.g. Creative Commons, MIT, BSD, Open Database License).

Domain shift¶

Training data originate from Radboudumc. Test data include cases from Radboudumc, CWZ, Karolinska Institute, and additional contributing institutes. Participants should expect domain shift between training and test cohorts due to differences in MRI acquisition parameters, H&E staining protocols, and scanner hardware. The organizers will match key clinical characteristics (e.g. age distribution, Gleason grade distribution) across cohorts as closely as possible.