Page 166 - Like me, or else... - Michelle Achterberg

Chapter 6
by using FLIRT (Jenkinson et al., 2002). The subject-specific seeds were then used to extract time series from preprocessed RS data.
First-level general linear models (GLM) were performed separately on time-series from each seed. The following nuisance signals were included: global signal, white matter (WM), cerebral spine fluid (CSF), 6 motion parameters and FD outliers. The global signal was included to reduce the influence of artifacts caused by physiological processes (i.e., cardiac and respiratory fluctuations) and scanner drifts (Birn et al., 2006; Fox and Raichle, 2007). In order to extract the time series for WM and CSF, we used subject specific WM and CSF masked, which were generated with FMRIB’s Automated Segmentation Tool (FAST, Zhang et al. (2001)). Additionally, each frame with an FD outlier, (FD>0.5 mm) was represented by a single regressor in the first-level GLM (see also Chai et al. (2014)). With this approach the amount of regressors is different between participants (ranging from 0-28). To account for this difference in first-level GLMs, the number of FD outliers (and thus the number of extra regressors) was added to the higher level statistical analyses as an additional covariate.
Figure 1. Subcortical seeds: ventral striatum (left), and amygdala (right). Higher-Level Seed Based Analysis
For both seeds, two higher-level group analyses were carried out using FMRIB’s Local Analysis of Mixed Effects (FLAME) stage 1; one for sample I and one for sample II. Higher-level analyses were performed using FLAME stage 1 with automatic outlier detection and included the number of extra regressors induced by the FD outlier modeling as covariate of no interest. Corrections for multiple comparisons were thresholded with Gaussian Random Field Theory cluster-wise correction with a minimal Z>3.09 (corresponding to p<.001) and cluster
164