Quality control strategies for brain MRI segmentation and parcellation: practical approaches and recommendations - insights from The Maastricht Study

Abstract

Background Quality control of brain segmentation is a fundamental step to ensure data quality. Manual quality control is the current gold standard, despite unfeasible in large neuroimaging samples. Several options for automated quality control have been proposed, providing potential time efficient and reproducible alternatives. However, those have never been compared side to side, which prevents to reach consensus in the appropriate QC strategy to use. This study aims to elucidate the changes manual editing of brain segmentations produce in morphological estimates, and to analyze and compare the effects of different quality control strategies in the reduction of the measurement error. Methods We used structural MR images from 259 participants of The Maastricht Study. Morphological estimates were automatically extracted using FreeSurfer 6.0. A subsample of the brain segmentations with inaccuracies was manually edited, and morphological estimates were compared before and after editing. In parallel, 11 quality control strategies were applied to the full sample. Those included: a manual strategy, manual-QC, in which images were visually inspected and manually edited; five automated strategies where outliers were excluded based on the tools MRIQC and Qoala-T, and the metrics morphological global measures, Euler numbers and Contrast-to-Noise ratio; and five semi-automated strategies, were the outliers detected through the mentioned tools and metrics were not excluded, but visually inspected and manually edited. We used a regression of morphological brain measures against age as a test case to compare the changes in relative unexplained variance that each quality control strategy produces, using the reduction of relative unexplained variance as a measure of increase in quality. Results Manually editing brain surfaces produced changes particularly high in subcortical brain volumes and moderate in cortical surface area, thickness and hippocampal volumes. The exclusion of outliers based on Euler numbers yielded a larger reduction of relative unexplained variance for measurements of cortical area, subcortical volumes and hippocampal subfields, while manual editing of brain segmentations performed best for cortical thickness. MRIQC produced a lower, but consistent for all types of measures, reduction in relative unexplained variance. Unexpectedly, the exclusion of outliers based on global morphological measures produced an increase of relative unexplained variance, potentially removing more morphological information than noise from the sample. Conclusion Overall, the automatic exclusion of outliers based on Euler numbers or MRIQC are reliable and time efficient quality control strategies that can be applied in large neuroimaging cohorts.