How a physicist uses it
- Join a challenge.Sign in and choose a released case with its prescription, structures, and scoring rules.
- Download the case.Receive CT and structure DICOM objects for planning in the participant’s own treatment-planning system.
- Create the best plan.Use Eclipse, Precision, Monaco, RayStation, or another TPS; the platform does not prescribe the optimizer.
- Upload RT Dose.Submit the resulting DICOM dose object while the service checks its relationship to the challenge case.
- Compare the result.Review DVHs, conformity, homogeneity, OAR metrics, isodose visualization, and live leaderboard position.
What actually happened
I built the platform and operated a real planning competition on it. Seven final plans completed the submission and evaluation workflow. Each was evaluated across 36 structures, and the anonymized final leaderboard ranged from 123.67 to 143.45 points out of 150. These figures come from the preserved final-evaluation logs; participant identities and institution names are deliberately omitted.
What I built to run it
The deployed system combines a React interface, FastAPI backend, PostgreSQL records, MinIO object storage, Redis/Celery evaluation jobs, Google login, interactive CT/structure/dose review, automated scoring, submissions, administration, and a real-time leaderboard. The private production repository is planchallengerepo; the public case study reports only aggregate, non-identifying outcomes.
The important boundary
This is an educational and benchmarking platform. Its scores compare submitted plans under challenge rules; they do not approve a clinical plan, verify treatment delivery, replace patient-specific QA, or establish that the highest-ranked plan is appropriate for an actual patient.
← All projects