GradLint validation suite: reproducibility code for b-vector flip detection, gradient-scheme QC, and auto-repair in diffusion MRI

Reproducibility code for the validation of GradLint, a fast, standalone quality-control gate for diffusion-MRI (dMRI) gradient tables. GradLint detects the silent b-vector flip / axis-permutation error, audits the gradient scheme and b-value structure, and writes an auto-repaired FSL bvec/bval and/or MRtrix .b table with a machine-readable provenance log.

This archive contains the analysis scripts that benchmark GradLint head-to-head against the field-reference checker, MRtrix3 dwigradcheck, across three multi-shell dMRI cohorts: HCP1200 (N=105, 1.25 mm), a single-site 3 T Siemens Prisma acquisition (N=57, 2 mm), and the OpenNeuro multiple-sclerosis dataset ds007908 (N=28, 1.5 mm). The user can use one or more of the three datasets for the analysis. 

The numbered stage scripts establish, end-to-end and antipode-aware: specificity (a correct acquired table is left untouched — no false repair), sensitivity/robustness (an injected flip is detected and correctly repaired down a severity ladder of Rician noise and direction-count reduction), agreement with dwigradcheck on the recovered convention, frame-aware FSL↔MRtrix cross-format emission on oblique/radiological affines, and per-subject runtime. Grading is done against a machine-readable ground-truth manifest. Input datasets are treated read-only; all output is written under a user-chosen base.

This suite is distributed separately from the installable GradLint package — it is the analysis code behind the published validation results, not part of the tool.

• GradLint source code: https://github.com/marcotag93/GradLint
• GradLint software archive (Zenodo): https://doi.org/10.5281/zenodo.20847595

Requires a release GradLint binary, MRtrix3 (dwigradcheck, mrinfo) on PATH, and Python ≥ 3.10. See the bundled README.md for the full protocol, input layout, and per-stage commands.

Licensed under MIT, matching GradLint.