The present study was approved by the Ethics committee of the Second Hospital of Anhui Medical University and all participants obtained written informed consent. Fifty hemodialysis ESRD patients (25 patients with SHPT, 14 males and 11 females, mean age 42.20±7.53 year (SHPT group); 25 patients without SHPT, 15 males and 10 females, mean age 41.96±6.17 year (Non-SHPT group)) from the nephrology department in Second hospital of Anhui Medical University between January 2017 and July 2018. All subjects were right-handed and complete the neuropsychological test. The inclusion criteria as following: (1) clinically diagnosed ESRD (estimated glomerular filtration rate eGFR less than 15 mL/min/1.73m2), (2) receiving regular hemodialysis three times a week at our hospital, and (3) more than 18 years older. Exclusion criteria were as following: (1) obviously encephalopathy revealed by clinician or image investigation, (2) neurologic complications of ESRD treatment, (3) any history of drug/alcohol abuse, (4) leukoencephalopathy on previous MRI, and (5) MRI contraindication.
In addition, 25 healthy right-handed controls (12 males and 13 females, mean age 41.80±7.15 year (HC group)) without history of neurologic, psychiatric, and traumatic diseases were recruited from the local community. All healthy controls had no kidney disease and did not perform laboratory test.
Before the MRI scan, we used Montreal Cognitive Assessment (MoCA) to assess the participants’ global cognitive abilities. It with a score less than 26 was diagnose of cognitive impairment.
Laboratory tests were detected in all patients to evaluate their renal function, serum creatinine, and urea levels within 24h prior to MRI scan.
MR imaging data were acquired with a 3.0T Siemens scanner in the Second Hospital of Anhui Medical University. All subjects were placed in a standard head coil. Foam padding was used to reduce head motion and scanner noise. We obtained 3D high resolution T1 weighted structural images (TE/TR=2.98/1900ms, FOV=256×256, slice thickness=1mm, voxel size=1×1×1mm3, number of slices=176) and diffusion weighted images (TE/TR=84/8400ms, FOV=256×256, slice thickness=3mm, slice gap=0, b=0,1000 s/mm2, direction=30) in all subjects. All subjects were told to remain still, keep their eyes closed and stay awake during the entirety of the MRI scan.
The data preprocessing and network construction were performed by PANDA (www.nitrc.org/projects/panda), which is a pipeline toolbox for diffusion MRI analysis. Briefly, the image preprocessing consisted of several steps: format conversion of original data (DICOM), BET (skull removal), eddy current and head motion correction, fractional anisotropy (FA) calculation, diffusion tensor tractography.
WM network construction
Network node definition. The structural images of each individual participant were firstly co-registered to their first b0 images and resliced into the DTI space by linear transformation. Then, the resliced structural images were non-linearly normalized to the MNI space. Finally, the derived deformation parameters were inverted and employed to warp the automated anatomical labeling (AAL) atlas from the MNI space to diffusion image native space. After this procedure, 90 cortical and subcortical regions (45 for each hemisphere) were obtained, each representing a network node (Table 1, Figure 1).
Network edge definition. To define the edges of structural network, we select a threshold value for the fiber bundles (with end-points in both nodes during the fiber tracking). To reduce false positive connections due to limited resolution of DTI when a minimum of three fibers as a threshold. The fiber number (FN) of the connected fibers between two regions as the weights of the network edges. Finally, the FN weighted structural networks were constructed for each participant, which was represented by a symmetric 90×90 matrix (Figure 1).
The WM network topological properties were analyzed using graph theory by GRETNA (www.nitrc.org/projects/gretna). To characterize the topological organization of WM structural network, several key measures were considered. They are the clustering coefficient (Cp), shortest path length (Lp), small-worldness (σ) (normalized clustering coefficient (γ) and normalized shortest path length (λ)), local efficiency (Eloc), and global efficiency (Eg). The Cp of a network is the average of the clustering coefficient of each node in the network, which indicates the local efficiency for the transformation of the information. The Lp of the entire network refers to the average shortest travel distance across all nodes and indicates the most efficiency information transfer between the two nodes.
To examine the γ and λ in this study, we generated 100 matched random networks which had the same number of nodes, edges, and degree distribution, but preserved the weighted distribution of the real network. We computed the γ (γ=Cp/Cprandom) and λ (λ=Lp/Lprandom), where Cprandom and Lprandom are the average clustering coefficient and shortest path length over the random network, respectively. A small world network should meet the γ is much larger than 1 and λ is close to 1.
Eg is the average of the inverse of the shortest path length of all node pairs in the network and usually reflects the ability of the network in parallel information processing. Eloc is the average of the global efficiency of the community neighboring all nodes in the network and represents the fault tolerance level of the network.
Identification of Hubs
Hubs of structural networks are essential nodes that are identified in various ways. In the present study, we applied regional efficiency to identify the hubs of the network. A node is considered as a hub if its regional efficiency is at least one SD greater than the mean nodal efficiency of the network.
Demographic and clinical data were analyzed using SPSS 23.0. Group differences in age, education and neuropsychological tests were used by one-way analysis of variance (ANOVA). Post hoc pairwise comparisons were then performed using t test. The gender data were analyzed by chi-square test. In addition, two-sample t test compared clinical data between two patients’ group.
For group effects in global and regional network metrics, comparisons were conducted among three groups by one-way ANOVA. Levels of significance were set at p<0.05 and a Bonferroni correction for multiple testing was applied for group differences (α=0.05/3=0.017). We also detected the relationship between network metrics and neuropsychological test scores and clinical index by correlation analysis. To investigate the correlation between neuropsychological test scores and specific brain regions, the nodes with significant group differences were taken.