Challenge Banner

Rules

Registration is closed and no new participants are being accepted. This page remains available for reference to participating teams and to anyone interested in how the Challenge was set up.

Participation and registration

  • Registration closed on the date given on the Timeline.
  • Participation is not anonymous. To qualify for evaluation, true names and affiliations [institute or company, country] must be displayed accurately on verified Grand Challenge profiles, for all participants.
  • Besides algorithms submitted by commercial parties, eligible open-source algorithms will be evaluated on the PRISM-AI dataset.

Algorithms and training data

  • No training data are provided. PRISM-AI is a benchmark of existing algorithms, not a model development competition; there is no development phase.
  • Eligible algorithms output a breast cancer probability score between 0.0 and 1.0 for one or more specified time periods. The probability of being diagnosed with breast cancer within 5 years is the primary outcome of the Challenge.
  • Submit the version of the algorithm you want evaluated. Algorithm versions and settings cannot be changed after the final submission deadline.

Submissions

  • This Challenge only supports the submission of fully automated methods in Docker containers. It is not possible to submit semi-automated or interactive methods.
  • All Docker containers submitted to the Challenge will be run in an offline setting: they will not have access to the internet and cannot download or upload any resources. All necessary resources (for example, pre-trained weights) must be encapsulated in the submitted container a priori.
  • Containers must respect the compute and runtime limits described in the Submission Guidelines: three minutes per case, with an optional NVIDIA T4 GPU (16 GB VRAM).

Results and publication

  • All algorithms are evaluated using a pre-defined Statistical Analysis Plan that was shared with participants and publicly available evaluation scripts on GitHub. The evaluation is identical for every participant.
  • Leaderboards are public and display algorithm names and affiliations. The leaderboards report the primary analysis: 5-year discrimination, calibration and clinical utility. Rankings are displayed per metric rather than as a single composite score, with discrimination displayed first. See Evaluation.
  • Secondary analyses are reported in the Challenge publication rather than on this platform.