Subjects: A total of 28 TAO patients and 22 HCs were recruited.
MRI acquisition: All subjects were examined by using a 3.0-T MR imaging system with a 20-channel head coil. High-resolution sagittal structural T1-weighted images were acquired using MPRAGE sequence, and functional images were then collected axially by an echo planar imaging sequence.
Data preprocessing: The rs-fMRI data were firstly preprocessed by using Data Processing Assistant for Resting-State fMRI advanced edition (DPARSFA) V4.4 (http://rfmri.org/DPARSF), and the preprocessing procedures included converting DICOM files to NIFTI images, removing the first 10 functional volumes, slice timing correction, realignment for head motion correction, reorientation, spatial normalization, spatial smoothing, nuisance covariates regression, and temporal band-pass filtering.
Static VMHC analysis: Static VMHC computations were then performed using DPARSFA V4.4. First, a mean image was created by averaging the normalized T1-weighted images for all participants. Second, this image was averaged with its left-right mirrored version to generate a group-specific symmetrical template. The normalized T1 images were then registered to the symmetric template and applied to the nonlinear transformation to the normalized functional images. Finally, for each participant, the homotopic connectivity was calculated as the Pearson’s correlation between the time series of each pair of mirrored interhemispheric voxels. Fisher r-to-z transformation was performed for the correlation coefficients to increase the normality of the distribution, and the VMHC z-maps were obtained.
Dynamic VMHC analysis: Dynamic VMHC were computed with the Temporal Dynamic Analysis (TDA) toolkit based on Data Processing & Analysis for Brain Imaging (DPABI) V3.1 (http://rfmri.org/DPABI), where sliding window-based analysis was applied to examine the whole-brain dynamic VMHC variability (sliding window, 32 TR [64 s]; step size, 1 TR [2 s]), and in each sliding window VMHC was calculated by the same method used in the computation of static VMHC. The standard deviation of z-values at each voxel of all windows was calculated to depict the dynamic VMHC.
Statistical analyses: Intergroup comparisons of static and dynamic VMHC images were conducted using SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12). Two-sample t test was performed to assess the group differences between TAOs and HCs, with age and gender controlled as confounding covariates. Statistical significance was based on a familywise error (FWE) correction for multiple comparisons at the cluster level (PFWE < 0.05) with a cluster-defining threshold of P < 0.001. The surviving brain regions were mapped onto the cortical surfaces using the BrainNet Viewer software package (http://www.nitrc.org/projects/bnv). The mean static or dynamic VMHC values in each significant cluster were extracted for each subject. After controlling the effect of age and gender, partial correlation analyses were performed to evaluate the relationships between static and dynamic VMHC values and clinical parameters in TAO group (P < 0.05).
SVM analyses: A linear SVM classifier was used to examine the performance of static and dynamic VMHC differences in distinguishing TAOs from HCs using the LIBSVM software (http://www.csie.ntu.edu.tw/~cjlin/libsvm/). Exploratory SVM analysis was conducted using a combination of these significant imaging features. Due to the limited number of samples, we employed a “leave-one-out” cross-validation (LOOCV) approach to evaluate the performance of the classifier. Receiver operating characteristic (ROC) curve analysis was used to examine the cross-validated performance of the SVM classification model. A non-parametric permutation test with 5000 permutations was applied to validate the significance of classification accuracy.