This paper studies how to check the quality of diffusion magnetic resonance imaging (dMRI) results at scale. The authors ran a structured visual quality control (QC) process on 18,328 dMRI scans from nine datasets. They looked at the outputs of seven processing pipelines that are representative of common dMRI analyses.

Previous work showed that simple summary numbers can miss common failure types. To investigate this, the team used a structured visual inspection that follows the full processing hierarchy from early steps to final outputs. In other words, reviewers did not just look at the final numbers; they looked at intermediate images and maps produced by each step in the pipeline.

A key finding is that a final output can appear acceptable while still depending on an earlier step that failed. Those upstream failures may only be visible when someone inspects the whole chain of outputs. This means that checking only end results or aggregate metrics can give a false sense of security about data quality.

The researchers also found that how finely QC should be applied depends on the algorithm. Some methods produce spatially structured outputs where a local problem affects only part of the image. Other methods produce outputs where a failure is global. That difference determines whether one can salvage part of a scan or must exclude it entirely.

The work shows that large-scale, structured visual QC for dMRI is feasible and useful. It argues for systematic checks that span all processing steps to keep quantitative findings valid and interpretable. Important caveats are that the study covered seven representative pipelines and nine datasets, so it may not capture every possible pipeline or data source. Also, structured visual QC requires human review and effort, and this paper demonstrates a needed approach rather than a one-size-fits-all solution.