Validation

SpinalfMRIprep is validated on multiple public and internal datasets spanning diverse acquisition protocols and clinical populations.

Validation Strategy

SpinalfMRIprep uses a two-tier validation approach:

Tier	Description	Purpose
Benchmark	1 subject × all sessions per dataset	Fast cross-dataset coverage for continuous testing
Full	All subjects × all sessions	Complete validation (in progress)

Benchmark Datasets

Benchmark validation runs 1 representative subject from each dataset with all available sessions.

Dataset	Source	Subjects	Sessions	Task	Status
ds005884	OpenNeuro	1	1	Motor	✅ Benchmark
ds005883	OpenNeuro	1	1	Pain	✅ Benchmark
ds004386	OpenNeuro	1	2	Rest	✅ Benchmark
ds004616	OpenNeuro	1	2	Hand Grasp	✅ Benchmark
Balgrist Motor	Internal	1	4	Motor	✅ Benchmark

Benchmark totals: 5 subjects, 10 subject-sessions

Full Validation Datasets

Full validation will process all selected subjects across all sessions.

Dataset	Source	Subjects	Sessions	Task	Status
ds005884	OpenNeuro	38	1	Motor	🔄 In Progress
ds005883	OpenNeuro	38	1	Pain	🔄 In Progress
ds004386	OpenNeuro	48	2	Rest	🔄 In Progress
ds004616	OpenNeuro	24	2	Hand Grasp	🔄 In Progress
Balgrist Motor	Internal	11	4	Motor	🔄 In Progress

Full validation totals: 159 subjects, 264 subject-sessions

Aggregate Metrics

Quality Control Outputs

Every preprocessing step generates:

1. qc_status.json - Machine-readable status (PASS/WARN/FAIL) with failure classification
2. Reportlets - Visual PNG figures for human inspection

Example QC Reportlet

QC reportlet examples will be added after initial validation runs.

Reproducibility

SpinalfMRIprep guarantees deterministic outputs:

• Containerized execution (Docker/Apptainer)
• Pinned dependency versions
• Seed-controlled randomization where applicable

---

For detailed QC specifications, see Reference → Schemas.