Edited by: Błażej Misiak, Wroclaw Medical University, Poland
Reviewed by: Chuanjun Zhuo, Tianjin Medical University General Hospital, China; Daniel Berge, Institut Hospital del Mar d'Investigacions Mèdiques (IMIM), Spain
This article was submitted to Schizophrenia, a section of the journal Frontiers in Psychiatry
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Schizophrenia is a heterogenous neuropsychiatric disorder with varying degrees of altered connectivity in a wide range of brain areas. Network analysis using graph theory allows researchers to integrate and quantify relationships between widespread changes in a network system. This study examined the organization of brain structural networks by applying diffusion MRI, probabilistic tractography, and network analysis to 48 schizophrenia patients and 24 healthy controls. T1-weighted MR images obtained from all participants were parcellated into 87 regions of interests (ROIs) according to a prior anatomical template and registered to diffusion-weighted images (DWI) of the same subjects. Probabilistic tractography was performed to obtain sets of white matter tracts between any two ROIs and determine the connection probabilities between them. Connectivity matrices were constructed using these estimated connectivity probabilities, and several network properties related to network effectiveness were calculated. Global efficiency, local efficiency, clustering coefficient, and mean connectivity strength were significantly lower in schizophrenia patients (
Schizophrenia is a debilitating mental disorder with an onset in early adulthood, a chronic course, and a considerable disease burden (
Various methods have been used to detect minute changes in brain structure and to analyze them in an integrated manner. Diffusion-weighted imaging (DWI) is a noninvasive method that displays parameters related to the diffusion of water molecules and can be used to provide diffusion tensor image (DTI) by measuring the diffusion of water in many different directions. The diffusion of water in the brain is restricted by cell membranes and cellular structures, and these limitations cause anisotropic diffusion or directional preference in diffusion. Thus, changes in diffusivity measured by DTI can reflect neural tissue damage or the organization of neural fibers. Rather than considering only the single dominant tensor orientation for each voxel, probabilistic tractography (
Network analysis is a branch of mathematics that describes a system as a graph, which is a set of nodes connected by a set of edges, and then analyses the topological characteristics of the graph (
Robustness is one of the network properties that may be analyzed; it is a concept developed through attempts to understand the brain's stability to physical damage. Robustness simulation, which involves quantifying changes in network properties after removal of components of the brain network, has revealed that the brain is more stable to random damage, but is more vulnerable to target deletion, which is the removal of specific regions in a particular order (
From the viewpoint of the distribution of connectivity, the normal brain network has a topological characteristic in which a relatively small number of regions (i.e., nodes constituting the network) are involved with the majority of connections (
The study of brain networks with diffusion MRI can facilitate the evaluation of the topologic characteristics of the entire brain structure, although it is difficult to accurately reflect the organization of complex brain fibers. Although several network studies have been conducted to date, definitive findings have not yet been achieved. Therefore, to expand the understanding of the structural characteristics of the brain in schizophrenia patients, this study aimed to assess the global network properties of the brain using extended multi-fiber probabilistic tractography. We reconstructed the brain network of each subject using diffusion imaging probabilistic tractography, and then we compared group differences in global network properties. In addition, we compared differences in nodal network properties and conducted a robustness simulation to better understand the stability and regional organization of the brain networks.
Subjects were recruited from the Asan Medical Center in Seoul, Korea. Forty-eight schizophrenia patients were enrolled, with all patients meeting the Diagnostic and Statistical Manual of Mental Disorders-IV-Text Revision (DSM-IV-TR) criteria. They were all right-handed, between the ages of 20 and 40 years old, and had no other known diseases that could affect brain function. They had all displayed psychotic symptoms such as delusions or hallucinations for <5 years. Twenty-four healthy controls who did not have any Axis I psychiatric diagnosis were enrolled. Furthermore, the healthy controls did not have any first-degree relatives with an Axis I psychiatric diagnosis. In addition, subjects were excluded if they were unable to complete MRI scanning sessions.
Written informed consent was obtained from all subjects. Ethnical approval for the study was obtained from the local Institutional Review Board, Asan Medical Center, Seoul, Korea.
MRI was performed on a 3-Tesla scanner with an eight channel SENSE head coil (Philips Achieva). Structural T1-weighted images were acquired with a turbo field echo sequence (FOV: 240 × 240 × 170 mm, voxel size: 1 × 1 × 1 mm, TE/TR: 4.6/4.9 ms). DWI images were acquired with an echo planar imaging (EPI) diffusion-weighted sequence. One baseline (b factor = 0 s/mm2) image and 32 diffusion-weighted gradient directions (b factor = 1,000 s/mm2) were acquired (FOV: 224 × 224 × 135 mm, voxel size: 2 × 2 × 3 mm, TE/TR: 70/5,422 ms, flip angle: 90°). Inappropriate images were found via visual inspection and excluded from further analyses.
Anisotropic voxel affects the distribution of anisotropic signal to noise ratios and can cause directional errors in the fiber tracking algorithm (
T1-weighted images were processed using the Desikan-Killiany atlas of FreeSurfer V 5.3 to parcellate discrete anatomical regions of interest (ROIs) (
To track white matter streamlines, we applied a probabilistic tractography method using the Diffusion Toolbox in the FMRIB software library (FSL) (
Brain networks were then reconstructed from the collection of ROIs and calculated connectivity probabilities, resulting in an association matrix. Each network was represented as a graph, G = (V, E), consisting of a set of nodes V (representing 87 ROIs) and connections E between the nodes (representing connectivity probability between nodes). To remove weak or spurious connections, we applied a threshold to the connectivity probabilities, with the lowest 10% of connections in each subject's network being discarded. Detailed information on the 87 ROIs used in the study is given in Supplementary Table
Several network properties of the reconstructed brain networks were evaluated to characterize their organization. These network properties were all based on the non-directional weighted matrices and were calculated using the MATLAB-based Brain Connectivity Toolbox (
The nodal local efficiency, degree, and betweenness centrality were calculated to describe the connectivity of the specific brain regions. Nodal local efficiency describes the inverse values of the shortest path length between direct neighbors of a given node, degree represents the number of all connections to a given node, and betweenness centrality is related to the number of shortest paths in a network passing through a given node. Nodal local efficiency quantifies the segregation properties of the network (
Global properties including the mean connectivity strength, global efficiency, clustering coefficient, local efficiency, and mean betweenness centrality were calculated to define the topological characteristics of the whole brain network. Mean connectivity strength is a global measure of the average connectivity probability values of all connections, and is related to the strength of the entire network (
Robustness is an indicator of network stability when brain damage is present (
Symptom severity was assessed using the Korean version of Positive and Negative Syndrome Scale (PANSS) score (
Demographic data were compared using independent
There were no significant differences in mean age or gender between the groups. The mean IQ scores and CTT-2 t scores were lower in patient group than in the healthy control group (
Demographic information on the schizophrenia (SPR) patients and healthy control subjects.
Age, years, mean ( |
28.9 (6.2) | 30.0 (5.3) | 1.66 | 0.418 |
Gender, male, |
19 (39.6) | 9 (37.5) | 0.29 | 0.864 |
FSIQ, mean ( |
97.8 (15.5) | 120.1 (9.2) | 7.86 | <0.001 |
CTT-t 1, mean ( |
48.6(14.4) | 54.5(7.5) | 1.88 | 0.064 |
CTT-t 2, mean ( |
47.0(13.7) | 63.8(20.5) | 4.15 | <0.001 |
GAF, mean ( |
39.8(19.3) | – | – | – |
PANSS, mean ( |
61.0(14.7) | – | – | – |
Patients with schizophrenia showed lower global efficiency, reduced local efficiency, reduced clustering coefficient, and reduced mean connectivity strength in comparison with the healthy control group. Mean betweenness centrality (245.1399 ± 10.2767 vs. 239.6925 ± 10.7912;
Global network properties of schizophrenia (SPR) patients and healthy control subjects.
Global efficiency | 1.14E-1 (2.26E-3) | 1.15E-1 (2.65E-3) | 0.47 | 0.042 |
Local efficiency | 1.02E-2 (7.87E-4) | 1.08E-2 (1.13E-3) | 3.22 | 0.011 |
Clustering coefficient | 7.64E-3 (6.42E-4) | 8.12E-3 (9.25E-4) | 3.20 | 0.013 |
Mean betweenness centrality | 245.14 (10.28) | 239.69 (10.80) | 0.01 | 0.041 |
Mean connectivity strength | 3.48E-1 (1.28E-3) | 3.55E-1 (1.63E-3) | 1.97 | 0.046 |
None of the findings reached the FDR-threshold. The patient group showed lower nodal local efficiency at an uncorrected (
The schizophrenia group showed increased nodal degree values in the left pars orbitalis, right lateral orbitofrontal cortex, right hippocampus, and right ventral diencephalon regions, while the healthy control group showed increased values in the right transverse temporal gyrus, right supramarginal gyrus, and right nucleus accumbens regions. The nodal betweenness centrality values in the right entorhinal cortex area of the schizophrenia patients were higher than in the healthy control group.
Table
Regions showing differences in nodal network properties between subject groups (uncorrected level of
Frontal | Left pars orbitalis | 4.26E-03 | 7.85E-04 | 4.90E-03 | 1.11E-03 | 0.012 |
Right caudal middle frontal gyrus | 1.15E-02 | 1.60E-03 | 1.24E-02 | 1.78E-03 | 0.049 |
|
Right medial orbitofrontal cortex | 1.10E-02 | 1.57E-03 | 1.23E-02 | 1.76E-03 | 0.006 |
|
Right pars opercularis | 8.87E-03 | 1.42E-03 | 9.86E-03 | 1.68E-03 | 0.007 |
|
Right precentral gyrus | 1.67E-02 | 2.61E-03 | 1.80E-02 | 2.76E-03 | 0.032 |
|
Right superior frontal gyrus | 2.12E-02 | 2.27E-03 | 2.25E-02 | 2.56E-03 | 0.022 |
|
Temporal | Left hippocampus | 9.04E-03 | 1.66E-03 | 1.00E-02 | 2.07E-03 | 0.220 |
Left superior temporal gyrus | 1.24E-02 | 1.44E-03 | 1.31E-02 | 1.35E-03 | 0.039 |
|
Left temporal pole | 6.95E-03 | 1.69E-03 | 7.99E-03 | 2.25E-03 | 0.045 |
|
Right hippocampus | 9.95E-03 | 1.55E-03 | 1.10E-02 | 1.72E-03 | 0.018 |
|
Right inferior temporal gyrus | 7.82E-03 | 1.12E-03 | 7.99E-03 | 1.23E-03 | 0.017 |
|
right superior temporal gyrus | 1.16E-02 | 1.53E-03 | 1.27E-02 | 1.50E-03 | 0.005 |
|
Right transverse temporal gyrus | 6.96E-03 | 1.03E-03 | 7.60E-03 | 1.22E-03 | 0.018 |
|
Parietal | Left precuneus | 9.74E-03 | 1.65E-03 | 1.04E-02 | 1.25E-03 | 0.024 |
Cingulate | Left caudal anterior cingulate cortex | 1.19E-02 | 1.94E-03 | 1.30E-02 | 2.06E-03 | 0.013 |
Left rostral anterior cingulate cortex | 1.22E-02 | 1.86E-03 | 1.33E-02 | 2.10E-03 | 0.019 |
|
Right rostral anterior cingulate cortex | 1.25E-02 | 1.88E-03 | 1.40E-02 | 2.33E-03 | 0.008 |
|
Basal ganglia | Left caudate nucleus | 1.27E-02 | 1.84E-03 | 1.41E-02 | 2.46E-03 | 0.033 |
Left nucleus accumbens | 8.37E-03 | 1.55E-03 | 9.44E-03 | 1.82E-03 | 0.025 |
|
Diencephalon | Left thalamus | 1.18E-02 | 1.76E-03 | 1.26E-02 | 1.84E-03 | 0.049 |
Left ventral diencephalon | 1.05E-02 | 1.02E-03 | 1.12E-02 | 1.47E-03 | 0.045 |
|
Frontal | Left pars orbitalis | 61.10 | 4.35 | 58.67 | 5.74 | 0.048 |
Right lateral orbitofrontal cortex | 78.40 | 3.22 | 76.58 | 2.75 | 0.007 |
|
Temporal | Right hippocampus | 83.06 | 2.45 | 82.79 | 1.06 | 0.030 |
Right transverse temporal gyrus | 77.60 | 3.47 | 80.04 | 2.16 | 0.003 |
|
Parietal | Right supramarginal gyrus | 80.85 | 2.32 | 82.00 | 1.67 | 0.050 |
Basal ganglia | Right nucleus accumbens | 76.98 | 4.14 | 79.13 | 2.88 | 0.033 |
Diencephalon | Right ventral diencephalon | 84.54 | 1.03 | 83.71 | 1.60 | 0.027 |
Parietal | Right entorhinal cortex | 44.83 | 38.99 | 23.92 | 22.13 | 0.025 |
Scatter plots of the number of deleted nodes and changes in network properties illustrate the resilience of the structural brain networks in the two study groups (Figure
Plots of robustness analysis in schizophrenia patients and healthy control subjects. In case of global efficiency, a linear mixed model was used to assess the group-by-number of removed nodes interaction.
In the patient group, there were significant positive correlations between CTT-1 t-score, and local efficiency and mean connectivity strength. The negative correlation of CTT-1 t-score with mean betweenness centrality was significant. CTT-2 t-score had a significant positive association with mean connectivity strength. In the patient group, there were no significant associations between IQ, GAF, and total PANSS score and network properties. In the control group, IQ raw score was significantly positively related to local efficiency, clustering coefficient, and mean connectivity strength.
Table
Relationship between network characteristics and clinical assessments.
Global efficiency | FSIQ (raw) | −0.09 | 0.569 | 0.33 | 0.138 |
FSIQ (adjusted) | 0.11 | 0.457 | −0.22 | 0.327 | |
CTT-t 1 | 0.30 | 0.053 | 0.41 | 0.062 | |
CTT-t 2 | 0.19 | 0.259 | 0.17 | 0.471 | |
PANSS | −0.08 | 0.598 | |||
GAF | −0.15 | 0.431 | |||
Local efficiency | FSIQ (raw) | 0.16 | 0.286 | 0.51 | 0.016 |
FSIQ (adjusted) | 0.20 | 0.174 | −0.10 | 0.642 | |
CTT-t 1 | 0.31 | 0.046 |
0.40 | 0.076 | |
CTT-t 2 | 0.31 | 0.056 | −0.14 | 0.539 | |
PANSS | 0.05 | 0.741 | |||
GAF | −0.14 | 0.468 | |||
Clustering coefficient | FSIQ (raw) | 0.18 | 0.238 | 0.52 | 0.014 |
FSIQ (adjusted) | 0.19 | 0.198 | −0.09 | 0.677 | |
CTT-t 1 | 0.30 | 0.056 | 0.39 | 0.081 | |
CTT-t 2 | 0.31 | 0.062 | −0.15 | 0.509 | |
PANSS | 0.07 | 0.675 | |||
GAF | −0.14 | 0.465 | |||
Mean betweenness centrality | FSIQ (raw) | −0.04 | 0.773 | −0.20 | 0.368 |
FSIQ (adjusted) | 0.01 | 0.942 | 0.32 | 0.152 | |
CTT-t 1 | −0.34 | 0.025 |
−0.23 | 0.309 | |
CTT-t 2 | −0.21 | 0.201 | 0.01 | 0.951 | |
PANSS | 0.07 | 0.656 | |||
GAF | 0.15 | 0.448 | |||
Mean connectivity strength | FSIQ (raw) | 0.11 | 0.471 | 0.49 | 0.020 |
FSIQ (adjusted) | 0.19 | 0.206 | −0.12 | 0.602 | |
CTT-t 1 | 0.43 | 0.005 |
0.35 | 0.120 | |
CTT-t 2 | 0.36 | 0.028 |
−0.13 | 0.579 | |
PANSS | 0.03 | 0.868 | |||
GAF | −0.13 | 0.517 |
The schizophrenia group showed significantly lower values than the control group in global efficiency, local efficiency, clustering coefficient, and mean connectivity strength. Global efficiency is an indicator of network integration, and reduced values could imply that the ability for functional integration across the overall network is degraded. The clustering coefficient and local efficiency value were both lower in the schizophrenia group, which means that the degree of segregation across the overall network, i.e., the local connectedness, was lower. In addition, decreased overall connectivity strength was also reported, suggesting that overall connectivity between regions in the schizophrenia group was different to that in the control group. In a previous study of brain anatomical networks in drug naïve schizophrenia patients, Zhang et al. suggested that decreased connectivity strength in subnetworks affects the deterioration of global topological characteristics (
As suggested in a previous report (
In comparison with the normal control subjects, the schizophrenia patients showed significantly lower nodal local efficiency values (at uncorrected
The nodal centrality indicators, as well as the mean betweenness centrality, showed results inconsistent with previous studies. While we found that the nodal degree value in the frontal and hippocampus areas and the betweenness centrality value in entorhinal cortex were rather increased in the schizophrenia group, previous studies found decreased degree in the frontal hub (
In the case of continuous network damage, a change in a parameter reflects the overall performance, such as the robustness and stability of the network (
A significant positive association between CTT-1 t-score and local efficiency and mean connectivity strength was found in the patient group, which is partly consistent with previous studies examining the relationship between processing speed measured by the verbal fluency test and the functional brain network (
However, the network properties showed no significant associations with GAF and PANSS total score or a significant relationship with PANSS positive and negative subscale scores in further analysis. These results are inconsistent with those of previous reports showing associations between symptom severity and reduced levels of overall connectivity and global efficiency (
Although structural connectivity is closely related to functional connectivity (
There are some limitations to this study which need to be addressed. First, as the diffusion imaging technique relies on water diffusion parameters and its spatial resolution is relatively low compared to the actual size of nerve fibers, diffusion MRI has difficulties with resolving complex fiber organizations, such as crossing, converging, diverging, and kissing fibers (
We compared brain structural networks between schizophrenia patients and healthy controls, using diffusion MRI probabilistic tractography and graph theory. When the topological network properties were compared, measures related to the global network integrity and segregation were significantly lower in the schizophrenia group. This suggests that schizophrenia could induce damage to the entire brain structure and deterioration of inter-regional connectivity, which results in less effective network organization. In addition, considering that group differences in nodal local efficiency were prominent in several regions, schizophrenia may be a disease characterized by network damage over a wide range of brain areas, although the damage disproportionally affects specific brain regions.
S-HS and JL: Conceptualization; S-HS, WY, and HK: Acquisition of data; S-HS, JL, SJ, and YK: Formal analysis; S-HS, JL, WY, HK, SJ and YK: Investigation; S-HS and JL: Original draft; All authors contributed to and approved the final manuscript.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This study was supported by the National Research Foundation of Korea (NRF-2012R1A1A1006514 and NRF-2017R1D1A1B03032707 to JL).
The Supplementary Material for this article can be found online at:
Regions of interests
diffusion-weighted images
diffusion tensor image
Markov chain Monte Carlo.