2.2. Data Preprocessing for fMRI Datasets

The preprocessing procedures for RS-EPI datasets were all carried out using AFNI (Analysis of Functional NeuroImages, https://afni.nimh.nih.gov/, accessed on 26 March 2021)’s suite of programs [41]. For preprocessing and artifact reduction of the RS-EPI data, the following steps were performed in sequence: truncation by initial ten volumes to yield steady-state magnetization, de-spiking (‘3dDespike’ in AFNI), physiological noise elimination including cardiac and respiratory artifacts (‘RETROICOR’ and ‘RVT’) [42,43], slice acquisition timing correction (‘3dTshift’ in AFNI), time-series alignment, Legendre polynomial detrending and motion- and hardware-related linear regressions [44,45,46], in order. For the ME dataset, a T₂^*-weighted echo combination was performed between the time-series alignment and the detrending and the regressions [4,38,47] by using AFNI (‘@compute_OC_weights’). All datasets were evaluated for artifacts due to abrupt head motion, passing the sudden motion detection of AFNI at the threshold level of 0.2 mm for the Euclidean L2 norm of motion displacement during each TR interval [45].

For functional connectivity (FC) estimation, the artifact-reduced residual time series were bandpass-filtered with a frequency range of 0.009 to 0.1 Hz and smoothed by a Gaussian kernel with a 4 mm full-width-at-half-maximum on volume dataset (‘3dTproject’ in AFNI). The preprocessed datasets were spatially normalized into the 2.0-mm-isotropic-resolution MNI template of ‘MNI152_T1_2009c+tlrc.HEAD’ by ‘auto_tlrc’ in AFNI. For FC investigation, the two ME datasets were concatenated after time-series z-score normalization as performed in the previous study [48] to render the final dataset with ~10-min length.