Ethics Statement
This prospective study was approved by an institutional review board (Peking University First Hospital Ethics committment) and was conducted in compliance with the Declaration of Helsinki. All participants in this study provided written informed consent.
Participants
Consecutive patients which were confirmed as PHN or LBP by anesthesiologists were included. All the subjects were right-handed. None of them had a history of psychiatric or neurological disorders. The clinical symptoms of pain were assessed using a numerical rating scales (NRS), with a range from 0 (no pain) to 10 (the highest tolerable pain). All of the patients underwent neurological and psychological examinations and fulfilled the mini-mental state examination (MMSE). Only the patients with the scores above 27 in MMSE were included. Patients with any abnormality detected on brain MRI were excluded from our study. Except for brain MRI, all the LBP patients also underwent lumbar MR scanning to identify the herniation of lumbar discs.
Data acquisition and preprocessing
All the brain MRI examinations were performed on a 3.0T scanner (GE healthcare, 750HD) with a 32-channel phased-array head coil. High-resolution structural T1WI were obtained by using a three-dimensional magnetization-prepared rapid gradient-echo (3D-MPRAGE) sequence with the following parameters: repetition time (TR) = 8.1 ms; echo time (TE) = 3.7 ms, flip angle = 80; slice thickness = 1 mm without a gap; field of view (FOV) = 240×240 mm2; matrix size = 256×256; voxel size = 0.86×0.86×1.0 mm3. A total of 160 axial slices were acquired for each patient.
All structural MRI data processing routines were carried out by using the Statistical Parametric Mapping 12 (SPM12, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) Toolbox in MATLAB-R2018b (The MathWork, Inc., Natick, MA, US). The preprocessing procedure was as follows. First, the MRI data of each subject were segmented into grey matter (GM), WM and cerebrospinal fluid (CSF). Second, the GM segments were non-linearly co-registered by using the inbuilt high dimensional Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) [5]. Third, GM images were normalized to standard Montreal Neurological Institute (MNI) space to make the GM images in the same space. Thereafter, the resulting GM images were modulated by the Jacobian determinants. Fourth, the GM images were smoothed with an 8 mm full-width-half-maximum (FWHM) Gaussian kernel. The GM images of all subjects were used for the further analysis.
We used the Automated Anatomical Labeling (AAL) atlas to define the whole brain parcellation [44]. The cerebellar regions were excluded due to incomplete coverage of the cerebellum in several participants. Thus, a total of 90 brain regions of interest (ROIs) were defined in this analysis.
GMV and MC estimation
In our study, we extracted two types of features using the preprocessed GM images: GMV and MC. GMV of each ROI was extracted as the local morphological feature. MC is a measure of structural connectivity, and it was calculated in the following steps. First, we quantified the GM intensity of each voxel within each ROI in preprocessed GM images. Second, the kernel density estimation with automatically chosen bandwidths [10] was used to calculate the probability density function for each ROI [11]. Third, the morphological connectivity for each pair of ROIs was estimated as the similarity between the two probability density functions of this pair of ROIs by using Kullback-Leibler divergence [23]. The Kullback-Leibler divergence was defined as:
![](https://myfiles.space/user_files/83062_751fab6dfaef2446/83062_custom_files/img1627372669.png)
where p(x) and q(x) were the probability density functions of two ROIs p and q, respectively.
Moreover, Kullback-Leibler divergence was converted to a similarity metric as [24]:
![](https://myfiles.space/user_files/83062_751fab6dfaef2446/83062_custom_files/img1627372681.png)
The range of Kullback-Leibler-based similarity (KLS) is from 0 to 1, where 1 indicates two identical distributions while 0 implies two completely different distributions. Finally, a MC matrix with a size of 90×90 was acquired for each participant.
Graph-theoretical network analysis
Next, we applied graph theory to estimate the network properties of MC, and the network properties including degree, small-worldness, network efficiency, clustering coefficient and characteristic path length. The calculation of network properties on MC matrices across all participants were performed with the GRETNA Toolbox [50]. First of all, a thresholding procedure is commonly used to binarize the MC network before performing topological characterization on the morphological connectivity matrices. A sparsity threshold, which was defined as the ratio of the number of existing edges divided by the maximum possible number of edges in a network, was used to binarize the divided MC network [23]. We binarized the MC network within a wide range of the sparsity threshold (from 0.05 to 0.4 with an interval of 0.02) because an automated method to determine the sparsity threshold is lacking [1,49]. Then, we calculated the network properties of MC within different sparsity thresholds.
Statistical analysis
Independent-t-test was used for the comparisons of age and pain scores between groups.
An independent two-sample t-test with an accompanying false discovery rate (FDR) correction was used to identify the brain regions where LBP and PHN patients showed significant GMV difference. Meanwhile, the total intracranial volume (TIV) of each subject was estimated and used as a covariate to remove the effect of variations in brain intracranial volume. Next, in order to explore the differences of MC network between LBP and PHN patients, we performed an independent two-sample t-test with FDR correction to correct the problem of multiple comparisons [8]. Furthermore, we detected the group differences in network properties (including degree, small-worldness, network efficiency, clustering coefficient, and characteristic path length) at different sparsity thresholds by using the two-sample t-test and FDR correction.