18K scans reveal hidden dMRI failures only visible through hierarchical QC
Downstream outputs can look perfect while upstream pipelines silently fail.
Researchers led by Michael Kim and Yihao Liu deployed a structured visual QC framework on 18,328 dMRI scans from nine datasets (including HABS-HD and ADNI). They evaluated outputs from seven conventional processing pipelines and performed hierarchical inspection of upstream and downstream dependencies. The study revealed that aggregate metrics can miss critical failure modes—downstream outputs may appear normal even when upstream steps (e.g., eddy current correction or registration) have failed.
Their analysis shows that appropriate QC granularity varies by algorithm: some failures require global exclusion of a scan, while others can be handled selectively. The team emphasizes that systematic, visual QC across the full pipeline hierarchy is essential for large-scale studies. Without such checks, quantitative findings may be invalid or misinterpreted. This work provides a practical roadmap for implementing structured QC in any large dMRI study.
- Structured visual QC applied to 18,328 scans across 9 datasets, evaluating 7 processing pipelines.
- Downstream outputs passing visual checks often rely on failed upstream dependencies, undetectable without hierarchical inspection.
- QC granularity must be algorithm-specific—some failures require global exclusion, others selective removal.
Why It Matters
Ensures validity of massive dMRI studies by catching silent failures that aggregate metrics miss.