FMRI Preprocessing

The neuroimaging data were preprocessed using SPM12 (Ashburner, 2012)¹. First, we defined the origin of each image to align with the anterior and posterior commissure plane (Ardekani and Bachman, 2009). After we motion-corrected each time series to its first volume, we then performed spatial unwarping to minimize geometric distortions due to susceptibility artifacts (Andersson et al., 2001; Hutton et al., 2002). Next, we coregistered the mean functional image to the anatomical scan and normalized the anatomical using the unified segmentation model (Ashburner and Friston, 2005). The normalized anatomical was subsequently used to reslice the functional data to standard stereotaxic space defined by the Montreal Neurological Institute (MNI). We applied a spatial smoothing at full-width half-maximize of 6 mm to the normalized functional data.

To minimize the impact of head motion on the neuroimaging data, we applied additional preprocessing steps using tools from FSL (FMRIB Software Library version 5.0.4; Smith et al., 2004)². We detected motion spikes using the FSL tools fsl_motion_outliers. The motion spikes were evaluated with two metrics: (1) root-mean-square (RMS) intensity difference of each volume relative to the reference volume obtained from the first time point; and (2) frame-wise displacements calculated as the mean RMS change in rotation/translation parameters relative to the same reference volume. We subjected the metric values within a run to a boxplot threshold (75th percentile plus 1.5 times the interquartile range) and labeled volumes as spikes, which were subsequently removed via regression (Satterthwaite et al., 2013; Power et al., 2015). Across all participants, this method removed 6.2% of volumes (range: 0.99–13.6%). After the removal of motion spikes, no participants exhibited extreme average volume-to-volume head motion (M = 0.058 mm; range: 0.027–0.10 mm) or maximum volume-to-volume head motion (M = 0.13mm; range: 0.060–0.26 mm). Following the removal of motion spikes, we extracted brain material from the functional images (Smith, 2002) and normalized the entire 4D dataset using a single scaling factor (grand-mean intensity scaling). Images were then processed through the SUSAN (Smallest Univalue Segment Assimilating Nucleus) noise reduction filter, part of the FSL software package, using a 2 mm kernel (Smith and Brady, 1997). This step allowed us to achieve greater signal-to-noise ratio while preserving the image structure. Lastly, we applied a high-pass temporal filter with a 100 s cutoff (Gaussian-weighted least-squares straight line fitting, with sigma = 50 s) to remove low frequency drift in the MR signal. Applying the temporal filter after the removal of motion spikes helps to minimize ringing artifacts (Weissenbacher et al., 2009; Carp, 2013; Satterthwaite et al., 2013).