MRI Data Acquisition and Preprocessing

The rs-fMRI images and T1-weighted MRI images were acquired using a 3.0 T MR imaging system (GE Healthcare, Discovery MR750, Milwaukee, WI, United States) in the Radiology Department of China-Japan Friendship Hospital. The parameters of sagittal three-dimensional T1-weighted images with fast spoiled gradient-echo sequences (FSPGR) were as follows: echo time (TE) = 3.0 ms, repetition time (TR) = 6.9 ms, slice thickness = 1.0 mm, FOV = 256 mm × 256 mm, acquisition matrix = 256 × 256, and flip angle = 12°. The parameters of axial resting-state data were as follows: TE = 30 ms, TR = 2,000 ms, slice thickness = 3.0 mm, 33 slices, field of view (FOV) = 240 mm × 240 mm, in plane matrix = 64 × 64, flip angle = 90°, and 240 phases.

Structural three-dimensional (3-D) T1 images were first processed using the SUIT toolbox¹ implemented in the Statistical Parametric Mapping software version 12 (SPM12)² toolbox (Diedrichsen et al., 2009). Each cerebellum was separated by a Bayesian algorithm into gray matter (GM) and white matter (WM), normalized to the Montreal Neurological Institute (MNI) space using the high-resolution probability template in SUIT. The intensity of each voxel was modulated to conserve the regional differences in the total amount of GM. All the images were smoothed with a 4-mm full-width at half-maximum (FWHM) Gaussian kernel.

The rs-fMRIs were preprocessed with the Data Processing Assistant for Resting-State fMRI (DPARSF) and the Resting-State fMRI Data Analysis Toolkit (REST). First, the first 10 volumes were discarded for the signal equilibrium and adaptation of subjects to the scanning noise. The remaining 230 volumes were corrected for timing difference and realigned to the first volume to correct for possible movement. The frame-wise displacement (FD) (Jenkinson) was calculated to evaluate the mismatch of volume-to-volume superimposed head position. The mean FD for the all the participants were 0.21 ± 0.14 mm. The data of 6 subjects (4 AD, and 2 NC) were excluded in this step due to excessive head motion (greater than 2.5 mm, greater than 2.5° angular rotation or mean FD > mean FD + 2SD). After removing the 6 subjects, the FD showed no significant different among the different groups (AD: 0.19 ± 0.09 mm, aMCI: 0.18 ± 012 mm; NC: 0.19 ± 0.09 mm, F-value estimated by one-way ANOVA was 0.45, p = 0.64). To normalize the resting images, the T1 images were registered to their corresponding functional images and were then segmented into GM, WM, and cerebrospinal fluid tissue (CSF) probabilistic maps using a unified segmentation algorithm. Second, a GM population template was derived from the whole image data set with the DARTEL technique. Third, non-linear warping of the segmented images was then performed to match the MNI space DARTEL template. Spatial smoothing was then performed with an isotropic 4-mm FWHM Gaussian kernel. Next, linear detrending and temporal band-pass filtering (0.01–0.1 Hz) were applied to remove low-frequency drifts and high-frequency noise. Finally, the nuisance variables (including 6 head motion parameters and their derivatives, the WM and CSF signal, and the linear term) were regressed out.