MRI preprocessing

In this study, we used a popularly-used fMRI preprocessing routine, as developed in the Data Processing Assistant for Resting-State fMRI (DPABI, http://rfmri.org/dpabi) [27, 28] and based on some functions in Statistical Parametric Mapping (SPM8, https://www.fil.ion.ucl.ac.uk/spm) [29]. All the preprocessing steps of T1-weighted and resting-state fMRI data were conducted by DPABI. The preprocessing pipeline was as follows. The first five volumes were removed to avoid a T1-equilibration effect, after which 175 volumes remained. The fMRI data consisted of images acquired one slice at a time; thus, each slice was acquired at a slightly different time point. Additionally, motion correction was used to adjust the time series of images so that the brain was in the same position in every image. Hence, we used DPABI to correct for differences in image acquisition time and head position from different slices by calling functions in the SPM. The timings of all slices were matched against the middle slice to ensure timing synchronization. The position of the head in each slice was adjusted to that in the first slice to ensure a fixed position across slices. Additionally, head motion parameters were obtained. The brain size, shape, orientation, and gyral anatomy varied largely across the participants. To enable inter-subject comparisons, MRI slices from each brain were transformed or spatially normalized into a standardized template [30]. The Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) function [31] in DAPBI was used to transform the functional data from the individual native space to the Montreal Neurological Institute space, and the functional data were resliced (3 × 3 × 3 mm³ voxels) and smoothed with a 4-mm FWHM. We further reduced the effects of physiological artifacts of whole-brain signals via a regression analysis in DPABI. In addition to the global mean signal, six motion parameters, cerebrospinal-fluid signals, and white-matter signals were removed as nuisance variables to reduce the effects of head motion and non-neuronal BOLD fluctuations. Before estimating dFC, temporal band-pass filtering (0.01–0.10 Hz) was performed to remove the effects of low-frequency drift and high-frequency noise in DPABI. The choice of ROIs determines the tradeoff between spatial coverage and resolution and should be carefully made. We chose Dosenbach’s ROIs, which are functionally representative to sample the whole brain [32]. Dosenbach’s ROIs have a clear coordinate definition for the location of structural partitions of the whole cerebral cortex and groups the ROIs into six types of networks, namely, the cerebellar, opercular, default, parietal, occipital, and sensorimotor networks. We also added four subcortical ROIs located in the bilateral amygdala and para-hippocampi according to previous studies [33], and these four ROIs were used as additional networks. Hence, we defined a total of 164 ROIs (spheres with a radius of 8 mm each), consisting of seven networks for subsequent whole-brain analysis. Then, we extracted the time series of each ROI by averaging the time courses of all voxels within the ROI. Finally, we divided the whole brain into seven networks: cerebellar, opercular, default, parietal, occipital, sensorimotor, and additional networks.