The benchmark evaluates machine learning methods for accelerated MRI reconstruction using hidden reference targets. Every challenge uses the same secure upload and server-side evaluation pipeline.
This track evaluates accelerated reconstruction at 8× acceleration.
Methods are ranked by PSNR, SSIM, and NMSE on hidden test targets, with optional anatomy-specific analysis.
Use multicoil_train/ and multicoil_val/ for model development, and submit predictions for
multicoil_test_full.
Maximum upload size: 8 GB per submission ZIP.
queued → evaluating → completed) and monitor leaderboard ranking.Use the challenge-specific notebook before uploading. It checks data structure, shape mismatches, naming consistency, and reports final ZIP size.
jupyter notebook validate_multicoil_test_reconstructions.ipynbWe strongly recommend running the notebook checks before every submission. benchmark validation tools
| Rank | Team Name | Submitted | Submission ID | |||
|---|---|---|---|---|---|---|
| Loading... | ||||||
This track studies cross-anatomy domain transfer. Models are developed without access to the held-out anatomy,
and are evaluated on hidden test reconstructions from that anatomy to measure out-of-domain generalization.
Submit predictions for files in anatomy_transfer_challenge/ankle/ corresponding to the
ankle joint (20 files, 49.40 GiB). You may have relatively few ankle volumes, so leveraging complementary
anatomies during training can improve performance.
This track evaluates held-out anatomy transfer. Maximum upload size: 500 MB per submission ZIP.
queued → evaluating → completed) and compare leaderboard metrics.Use the challenge-specific notebook before uploading. It checks data structure, shape mismatches, naming consistency, and reports final ZIP size.
jupyter notebook validate_ankle_challenge_reconstructions.ipynbWe strongly recommend running the notebook checks before every submission. benchmark validation tools
| Rank | Team Name | Submitted | Submission ID | |||
|---|---|---|---|---|---|---|
| Loading... | ||||||
This track targets contrast-domain shift and evaluates reconstruction on held-out T1-weighted fat-suppressed (T1-FS)
acquisitions. Leaderboards report global and anatomy-filtered PSNR, SSIM, and NMSE.
Submit predictions for files in contrast_generalization_challenge/T1_FS/.
T1-FS examples can be limited (17 files, 20.74 GiB), so mixing additional anatomies/contrasts during training is often beneficial.
This track targets protocol shift in T1-weighted fat-suppressed (T1-FS) acquisitions. Maximum upload size: 700 MB per submission ZIP.
queued → evaluating → completed) and compare leaderboard metrics.Use the challenge-specific notebook before uploading. It checks data structure, shape mismatches, naming consistency, and reports final ZIP size.
jupyter notebook validate_contrast_challenge_reconstructions.ipynbWe strongly recommend running the notebook checks before every submission. benchmark validation tools
| Rank | Team Name | Submitted | Submission ID | |||
|---|---|---|---|---|---|---|
| Loading... | ||||||