AUTOMATED QC OF DWI DATA

In 2019, the Kimel lab formed a DWI working group. One of its tasks was to create a method of automatically performing QC on large DWI datasets, as a complement to visual review. Ultimately, we found very high agreement between trained human raters and 6 quality metrics derived from eddy and MRTrix. The method was tested on 3 large datasets (SPINS, POND, and a subset of HCP) and compared to ratings of 5 independent human raters.

The metrics, and their suggested thresholds, were as follows:

Metric	Absolute motion	Relative motion	Percent outliers	Average SNR	Average CNR	Residual noise
Suggested threshold	>= 2 (mm)	>= .5 (mm)	>= 2 (%)	<= 20	<= 1.5	>= 6

---

Repo Contents

This repo contains code to format and review the 6 requiste metrics, produce a score for each participant that can be used as a quality regressor in future analyses, and indicate which participants should likely be removed from further analyses. Specifically:

code

• 01_qualityScore.Rmd. This script performs two tasks. First, it examines the correlations between the 6 QC metrics. The intent is to ensure that, on a global level, there are no unexpected associations between quality metrics. Second, it performs PCA. The intent to to generate a 'quality regressor' that can be used in subsequent analyses, assuming the first one or two PCs capture a large proportion of variance.
• 02_excludeParticipants.Rmd. This script visualizes participant scores on the 6 QC metrics, and summarizes which participants should be excluded on the basis of the groups' established thresholds.

/data

• /input

  • eddyMetrics.tsv. This .tsv is generated by lab pipelines for data found in the /archive. Specifically, it can be found in bids/pipelines/dwiprep/.
  • mrtrixResidualNoise.txt. This output should be generated by the user (it does not, at present, exist in the /archive. A script to do so can be found at here (currently private).

• /output.

  • df_dwiQC_PC.csv is generated by 01_qualityScore.Rmd. It is a dataframe containing the 6 variables of interest, as well as the first PC score.
  • excludeParticipants.txtis generated by 02_excludeParticipants.Rmd. It contains the IDs of those participants exceeding exclusion thresholds.

Examples.
When knitted in RStudio, 01_qualityScore.Rmd should generate a report similar to the one here, and 02_excludeParticipants.Rmd as here. For more context, notes from the DWI working group meetings here