Page 205 - Like me, or else... - Michelle Achterberg

MRI scan quantity and quality in childhood
203
MRI data quality control
Motion estimation of functional MRI (task-based and resting state) was carried out using Motion Correction FMRIB’s Linear Image Registration Tool (MCFLIRT Jenkinson et al. (2002), as implemented in the FMRIB Software Library (FSL) version 5.09 (Smith et al., 2004). Absolute displacement (AD) in x, y, and z direction was calculated for all runs, for all participants (Table 1), with the middle volume of the run as a reference. We additionally investigated micro-movement (i.e., motion between two volumes) using the motion outlier tool (fsl_motion_outliers). Mean framewise displacement (FD) was calculated for all runs, for all participants (Table 1). Reliability analyses showed consistency in head motion over fMRI runs: mean FD: α=.77; mean AD (mean x-y-z direction): α=.84. For further analyses we computed a mean score over all fMRI runs for framewise displacement (M=.77, SD=1.29, range=.09-17.5) and absolute displacement (M=2.55, SD=3.77, range=.21-37.91). Framewise and absolute displacement were significantly positively correlated: r=.88, p<.001. For task- based fMRI runs, we defined runs with <3 mm (1 voxel) maximum motion in all directions as sufficient quality (Achterberg et al., 2018b; van der Meulen et al., 2018). For the RS fMRI data, volumes with framewise displacement of >0.3 mm (stringent threshold) or >0.5 (lenient threshold) were flagged as outliers (Power et al., 2012). RS fMRI data with < 20% of the volumes flagged as outlier was classified as sufficient quality, see Table 1. Although inclusion criteria for task- based and RS fMRI were different, they resulted in comparable motion estimates for the different fMRI runs of included participants (Table 1).
Structural T1 scans were pre-processed in FreeSurfer (v5.3.0). Anatomical labeling and tissue classification was performed on the basis of the T1- weighted MRI image using various tools of the FreeSurfer software (http://surfer.nmr.mgh.harvard.edu/). The pre-processing pipeline included non-brain tissue removal, cortical surface reconstruction, subcortical segmentation, and cortical parcellation (Dale et al., 1999; Fischl et al., 1999). After pre-processing, each scan was manually checked to assess quality by three trained raters. Scans were rated based on a set of specific criteria (e.g., affection by movement, missing brain areas in reconstruction, inclusion of dura or skull in reconstruction, see Klapwijk et al. (2019). 31% of the structural T1 scans were rated as ‘Excellent’, 43% of the scans were rated as ‘Good’, 16% of the scans were rated as ‘Doubtful’, and 10% of the scans were rated as ‘Failed’ (see Figure 2a). Structural anatomical data rated as ‘Failed’ and ‘Doubtful’ were classified as insufficient quality, and data coded as ‘Excellent’ and ‘Good’ were classified as sufficient quality. We investigated whether scans with different ratings would show actual differences in estimated brain volume, by comparing the four different ratings on the “Total Gray Volume” variable from the Freesurfer output. We found a significant difference in gray matter volume between the different ratings (F(3, 463) = 5.07, p = .002), with post hoc analyses revealing a significant difference between scans rated as ‘Failed’ and scans rated as ‘Excellent’ to
7