Table 1.

Quality control approaches for sMRI data

Method	QC input metrics	Visual QC/ classifier categories	Technique	QC output	Performance
FD (Savalia et al., 2017)	FD from functional MRI scan of the same session as proxy for motion in T1-weighted images	Three categories: pass, warn, fail	Flagging procedure; combining visual QC and estimates of head motion from functional MRI scans	FD estimates and visual QC ratings	FD estimates complement visual QC rating
Euler number (Rosen et al., 2018)	Euler number outputted by FreeSurfer	Three categories: 0 (gross artifacts/fail), 1 (some artifacts but usable), 2 (no artifacts)	/	Euler number, no specific recommendations	Euler number as most accurate quality measure/highest correlation with visual QC
MRI-QC (Esteban et al., 2017)	Raw T1-weighted images, 64 IQMs per input image	Binary classifier: include, exclude	random forests classifier trained on a publicly available, multi-site data set (17 sites, N = 1102)	individual anatomical reports (calculated IQMs and metadata in the summary, as well as a series of image mosaics and plots designed for the visual assessment of images)	Intra-site prediction: high accuracy; Unseen site prediction: leaves space for improvement (76 % ± 13 % accuracy)
Qoala-T (Klapwijk et al., 2019a, 2019b)	Metrics form the FreeSurfer output files aseg.txt, aparc_area.txt and aparc_thickness.txt (all for both hemispheres) including the variable surface holes	Four categories: 1 (excellent), 2 (good), 3 (poor), 4 (failed)	supervised-learning model, random forests classifier trained on the BrainTime data	Qoala-T score (ranging from 0 to 100), recommendation whether to visually check and whether to include or exclude each data set from further analyses	Intra-site prediction: high accuracy (mean AUC = 0.98); Unseen site prediction: similar accuracy (mean AUC = 0.95)

AUC area under the curve, QC quality control, FD frame-by-frame displacements, IQM image quality metrics, MRI magnetic resonance imaging