QC reports

Every run of SpinePrep produces self-contained HTML QC reports (no server needed) — the human-facing layer of the "visual QC is the validator" philosophy. There are two, mirroring the BIDS-App levels.

Subject report (participant level)

derivatives/.../sub-XX/sub-XX_qc_report.html — one per subject. Sections:

1. Summary card — an Include / Review / Exclude recommendation with a one-line reason, headline mean FD and median in-cord tSNR, and a step strip (S1→S10 PASS/WARN/FAIL chips).
2. Anatomical — cord segmentation + vertebral labelling and PAM50 registration.
3. Per-run functional cards — for each BOLD run: per-step chips, embedded headline reportlets (S4 motion, S5 distortion before/after, S9 tSNR), a metric micro-table (FD, % censored, cord-Dice, A-P displacement, condition #, tSNR pre→post), and a structured what-failed-and-why note for non-PASS runs (including the honest "distortion-limited, no fieldmap" label).
4. Confound model — the nuisance regressors emitted for your GLM.
5. Methods boilerplate (auto-generated, CC0) + a provenance footer.

Group report (group level, per dataset)

group_report_<dataset>.html + the cross-dataset release_report.html. Sections: cohort summary; an inclusion table (per-subject recommendation, shaded + linked); MRIQC-style QC-metric distributions with literature reference lines; per-vertebral-level views; a CONSORT-style attrition waterfall (every drop reconciles to the prior step's FAILs — no silent losses); WARN/FAIL stratification; and a reproducibility panel.

Example figures (from the validation cohort)

These are produced by the same machinery that feeds the reports:

A fully interactive gallery (live subject + group reports embedded) ships with the hosted site; the report HTML lives in each run's derivatives/.../release/.