QSM-CI

An open benchmarking platform and leaderboard for Quantitative Susceptibility Mapping (QSM) reconstruction.

Submit a QSM algorithm — in any language, as a container — and QSM-CI runs it on standardized data, scores it against held-out ground truth, and publishes the result to an interactive leaderboard. Every published algorithm gets a citable Zenodo DOI and can be run by anyone with one command.

→ https://qsmxt.github.io/QSM-CI

• Results — per-stage tables, the background-removal × dipole-inversion combination matrix, and findings, with an interactive volume viewer.
• Run an algorithm — locally or from a workflow engine (nipype, Pydra, CWL, Snakemake, Nextflow).
• Submit yours — open a pull request adding a folder under algorithms/.

Run an algorithm

pip install qsm-ci
qsm-ci list                                   # the published algorithms you can run
qsm-ci run sharp --totalfield tf.nii.gz --mask mask.nii.gz

Algorithms are fetched from Zenodo and run in their pinned container, so results are reproducible. See the running guide for details.

Submit an algorithm

A submission is a folder under algorithms/<slug>/ — your run.sh, a stage: declaration, and a container environment — reading and writing canonical artifacts (all fields and χ in ppm). Your code is mounted, not baked. QSM is scored stage-aware (field-mapping → background removal → dipole inversion, plus combined spans), both isolated and composed with other stages.

Start from the Submit guide; the full rules are in CONTRACT.md and the machine-readable stages.yml. MATLAB is welcome via the license-free MATLAB Runtime.

Repository layout