The thalamic covariance network is associated with cognitive deficits in patients with cerebral small vascular disease

Abstract

Objective

Abnormalities in the gray matter structure of cerebral small vessel disease (CSVD) have been observed throughout the brain. However, whether cortico-cortical connections exist between regions of gray matter atrophy in patients with CSVD has not been fully elucidated. This question was tested by comparing the gray matter covariance networks in CSVD patients with and without cognitive impairment (CI).

Methods

We performed multivariate modeling of the gray matter volume measurements of 61 patients with CI (CSVD-CI), 85 patients without CI (CSVD-NC), and 108 healthy controls using source-based morphological analysis (SBM) to obtain gray matter structural covariance networks at the population level. Then, correlations between structural covariance networks and cognitive functions were analyzed in CSVD patients. Finally, a support vector machine (SVM) classifier was used with the gray matter covariance network as a classification feature to identify CI among the CSVD population.

Results

The results of the analysis of all the subjects showed that compared with healthy controls, the expression of the thalamic covariance network, cerebellum covariance network, and calcarine cortex covariance network was reduced in patients with CSVD. Moreover, CSVD-CI patients showed a significant reduction in the expression of the thalamic covariance network, encompassing the thalamus and the parahippocampal gyrus, relative to CSVD-NC patients, which persisted after excluding CSVD patients with thalamic lacunes. In patients with CSVD, cognitive functions were positively correlated with measures of the thalamic covariance network. More than 80% of CSVD patients with CI were correctly identified by the SVM classifier.

Interpretation

Our findings provide new evidence to explain the distribution state of gray matter reduction in CSVD patients, and the thalamic covariance network is the core region for early gray matter reduction during the development of CSVD disease, which is related to cognitive deficits. Reduced expression of thalamic covariance networks may provide a neuroimaging biomarker for the early identification of cognitive impairment in CSVD patients.

Introduction

Cerebral small blood vessel disease (CSVD) refers to local damage to brain tissue caused by small blood vessel lesions in the brain from a variety of causes.¹ The incidence of CSVD is associated with age, affecting approximately 5% of those aged 50 to nearly 100% of those older than 90.² The main consequences of CSVD are cognitive deficits of varying degrees, including dementia.^3,4

Structural magnetic resonance imaging has been widely used in the study of morphological differences in the brains of patients with CSVD. Previous studies have shown gray matter atrophy in multiple regions, such as smaller frontal cortex volumes and a thinner precuneus, that have been shown to be associated with cognitive decline.^5–9 Most previous studies on CSVD have been based on voxels or vertices, comparing the mean differences between groups between all voxels or vertices in the brain of interest without considering any spatial information across voxels/vertices.

Previous studies have shown that the morphological characteristics of different brain regions covary in populations; moreover, gray matter structural covariance networks are obtained by converting at the population voxel/vertice level, which provides information about localization of gray matter changes and their variation among individuals.¹⁰ In addition, gray matter structural covariance networks partially overlap with known functional networks and correlate with behavioral and cognitive ability.^11,12

Source-based morphological analysis methods (SBM) can cluster voxels/vertices with similar information to construct gray matter structural covariance networks among subjects.¹⁰ Compared with traditional univariate measurement methods for brain morphology research, SBM mainly has the following two advantages: (1) the region of interest does not need to be defined in advance and (2) noise can be removed with high precision.^10,13 The SBM approach has been widely used in neurological and psychiatric studies to build structural covariance networks, ultimately helping to reveal differences in brain structural covariance networks in healthy, neurological, and psychiatric populations, such as patients with multiple sclerosis,¹⁴ depression,¹⁵ schizophrenia,¹⁶ etc.

A previous study applying the SBM approach in older CSVD patients with CI (CSVD-CI patients; over 75 years of age) revealed abnormal gray matter structural covariance networks that were associated with cognitive deficits.¹⁷ However, among people older than 50, the incidence of CSVD reaches 5%, and it increases exponentially with age.¹⁸ CSVD patients can experience varying degrees of cognitive decline, and early intervention can help delay cognitive decline. In addition, to our knowledge, no studies have investigated differences in structural covariance networks between CSVD-CI patients and CSVD patients without CI (CSVD-NC patients) by utilizing the SBM approach.

The primary purpose of this study was to investigate structural covariance network differences between CSVD-CI and CSVD-NC patients (50–75 years old). We hypothesized that cognitive decline is prevalent in CSVD patients older than 50 years, there are differences in structural covariance network expression between CSVD-CI and CSVD-NC patients, and that cognitive decline is related to cognitive defects; thus, these findings may help to establish neuroimaging markers to distinguish CSVD-CI from CSVD-NC patients in the early stage, improving clinical diagnosis and treatment decision-making.

Materials and Methods Participants

This study was approved by our hospital's ethics committee. We recruited patients admitted to the Department of Neurology, the Affiliated Hospital, North Sichuan Medical College, who received brain MRI and were diagnosed with CSVD between September 2016 and September 2020. The following criteria were used to define CSVD¹⁸: (1) white matter hyperintensities (WMHs): “Signal abnormality of variable size in the white matter that shows the following characteristics: hyperintensity on T2-weighted images such as fluid-attenuated inversion recovery, without cavitation (signal different from CSF). Lesions in the subcortical gray matter or brainstem are not included in this category unless explicitly stated. If deep gray matter and brainstem hyperintensities are also included, the collective term should be subcortical hyperintensities. (2) lacunar infarction (LI): A round or ovoid, subcortical, fluid-filled cavity (signal similar to CSF) of between 3 mm and approximately 15 mm in diameter, consistent with a previous acute small subcortical infarct or hemorrhage in the territory of one perforating arteriole”.¹⁸ Patients with CSVD were divided into two groups: those without cognitive deficits (CSVD-NC group) and those with cognitive impairments (CSVD-CI group). The inclusion criteria of the CSVD-NC group were as follows: (1) met the diagnostic criteria of CSVD; (2) had no recent CI chief complaint and normal daily life activities; and (3) according to the grouping criteria¹⁹ and considering regression age and years of education, did not have any two cognitive domains with regression-adjusted test values less than the mean of the healthy control group −1.5 SD. The inclusion criteria of the CSVD-CI group were as follows: (1) met the diagnostic criteria of CSVD; (2) according to the grouping criteria¹⁹ and considering regression age and years of education, had at least two cognitive domains with a regression-adjusted test value less than the mean of the healthy control group −1.5 SD. After a full explanation of the procedures involved, all subjects provided written informed consent. A total of 146 patients with CSVD and 108 healthy controls (HCs) were enrolled in this study.

Exclusion criteria for all participants included a history of brain trauma, mental or neurological disease, peripheral neuropathy, migraine, tremor, other preexisting brain lesions visible on magnetic resonance imaging (MRI) (other than WMH and LI), other medical complications, alcohol or illicit drug use, and pregnancy.

Clinical assessment

All participants underwent a comprehensive neuropsychological test that included the following components: (1) The general cognitive ability of each participant was assessed by the MMSE. (2) The memory test included evaluations of auditory, visual and working memory. The Chinese version of Rey's Auditory Verbal Learning Test (RAVLT) is widely used to assess auditory memory. Visual memory is often assessed with the Rey-Osterrieth Complex Figure Test (ROCF), and the backward Digital Span Test (B-DST) is used to test working memory. (3) The Trail Making Test, Part A (TMT-A) and Commissioning test, Part B (TMT-B) were also used. (4) Language ability was evaluated with the Boston Naming Test (BNT, 30 questions) and Language Fluency Test (VFT). (5) Visual–spatial function was assessed with the clock plotting test (CDT).

MRI acquisition

Magnetic resonance imaging scans were performed on a 32-channel head-phased front circle 3.0 T MRI scanner (Discovery MR750, USA). Foam pads and earplugs were used to limit head movement and minimize scanner noise, respectively. The parameters of high-resolution 3D-T1 were as follows: TR = 8.3 ms, TE = 3.3 ms, turning angle = 15°, thickness = 1.0 mm, interval = 0 mm, field of view (FOV) = 240 mm × 240 mm, voxel size = 1 × 1 × 1 mm³. The matrix size was equal to 256 by 256. WMH and LI were observed on T2 fluid decay inversion recovery (T2-FLAIR)-weighted images. The parameters for T2-FLAIR-weighted images were TR = 8000 ms, TE = 126 ms, TI = 1500 ms, thickness = 3.0 mm, interval = 1 mm, FOV = 240 mm × 240 mm, and matrix size = 256 × 192.

Image preprocessing (using a morphological measurement method based on the voxel level)

The CAT12 software package in SPM was used to preprocess the acquired high-quality 3D-T1WI MRI images. Briefly, the original 3D-T1WI MRI images were divided into gray matter images and white matter images, and then the DARTEL algorithm and modulation were used.²⁰ Finally, an 8 mm × 8 mm × 8 mm Gaussian smoothing kernel FWHM was used to smooth the modulated gray matter image.

Construction of the gray matter structure covariance network (using the source-based morphological analysis method)

The gray matter image obtained from the image preprocessing in Step 1 was processed using the SBM program in the GIFT toolbox. The processing steps were as follows: (1) gray matter images of each subject were converted into one-dimensional vectors, (2) the one-dimensional vectors of all subjects were concatenated into a two-dimensional matrix (row: one-dimensional vectors of each subject × column: all voxels), and (3) this two-dimensional matrix was decomposed into a source matrix and a mixing matrix by independent component analysis. The columns in the mixing matrix represented spatial components, a group of regions called a “network,” the rows represented subjects, and the corresponding weight coefficients of each row represented the contributions of each subject to this component.²¹ (4) The minimum description length (MDL) function in the GIFT toolbox was used to automatically estimate the number of spatial components that could be clustered. (5) Independent component analysis was performed using Infomax.²² (6) ICASSO was used for 100 iterations to evaluate component stability. Following these steps, a total of 29 spatial components were generated. According to the criteria defined by Xu et al., 16 components were visually identified as obvious artifacts, such as those located at sharp edges, especially near cortical boundaries, or mainly appearing in areas without gray matter (such as white matter or ventricles). Therefore, a total of 13 components were analyzed in this study (supplementary materials S1 and S2; Table S1 for specific anatomical labels). (7) The source matrix was then visualized; the source matrix represented the relationship between spatial components (rows) and voxels (columns). Each row of the source matrix was reshaped into a three-dimensional brain graph and scaled according to unit standard deviation (Z graph), and the threshold was set as 3.0.¹⁰

Statistical analysis Gray matter structure covariance network differences: intergroup differences in weight coefficients

The weight coefficients of the 13 spatial components in the three groups were compared by one-way analysis of variance. Age, sex, and years of education were considered covariates. Post hoc t tests were subsequently applied to detect differences between each pair of groups. When p < 0.05, the Bonferroni method was used to correct the results.

Correlation analysis with cognitive function

All cognitive indicators were classified into five cognitive units by dimensionality reduction Z-transform normalization. Partial correlation analysis was conducted between the weight coefficients of the gray matter structure covariance network with significant differences between CSVD patient groups and the scores of each cognitive unit after Z-transform normalization. Age, sex, and years of education were considered covariates. When p < 0.05, the FDR method was used to correct the results.

Support vector machine (SVM) was used to build the classification model

The SVM comprised the following steps: (a) The weight coefficient of the gray matter structure covariance network that showed significant differences between the CSVD-CI and CSVD-NC groups (as analyzed above) was combined with the cognitive scale score (excluding the measures used for grouping CSVD patients) to establish the classification model as a training group, and then the cross-validation method was used to verify the model. (b) Receiver operating characteristic (ROC) curves were used to evaluate the model performance.

Result Demographic and cognitive assessments

As shown in Table 1, a total of 254 subjects older than 50 were included in this study, including 61 patients with CSVD-CI, 108 HC patients, and 85 CVSD-NC patients. As shown in Table 2, all cognitive indices of the CSVD-CI and CSVD-NC patients were lower than those of the HC group.

Table 1 Demographic and neuropsychological data for all subjects.

	CSVD-CI (n = 61)	HC (n = 108)	CSVD-NC (n = 85)	CSVD-CI vs. HC vs. CSVD-NC		CSVD-CI vs. HC	CSVD-CI vs. CSVD-NC	HC vs. CSVD-NC
				F/χ2	p	p	p	p
Sex (1/0)	25/36	66/42	36/49	9.305	0.010	0.879	0.012a	0.010c
Age	68.77 ± 5.95	67.31 ± 5.70	70.35 ± 4.73	4.888	0.008	0.177	0.161	0.002c
Education	9.59 ± 1.73	10.30 ± 2.24	10.20 ± 2.40	1.818	0.164	0.081	0.097	0.991
BMI	24.39 ± 6.57	23.60 ± 2.68	24.66 ± 5.52	1.215	0.298	0.319	0.735	0.136
HR	75.26 ± 14.27	75.19 ± 7.96	75.28 ± 13.00	0.002	0.998	0.967	0.992	0.954
SBP	147.37 ± 23.25	129.19 ± 14.25	144.28 ± 22.14	22.354	<0.001	<0.001b	0.347	<0.001c
DBP	79.42 ± 15.35	79.71 ± 9.44	79.71 ± 9.44	0.450	0.638	0.879	0.404	0.426
BG	5.83 ± 1.70	6.40 ± 2.99	6.09 ± 1.44	1.270	0.283	0.121	0.503	0.347
TC	4.45 ± 1.53	5.08 ± 1.14	4.15 ± 1.35	12.560	<0.001	0.003b	0.181	<0.001c
TG	1.37 ± 0.58	1.74 ± 1.68	1.49 ± 0.59	2.111	0.123	0.056	0.552	0.153
HDL	1.39 ± 0.42	1.51 ± 0.34	1.40 ± 0.39	2.697	0.069	0.056	0.888	0.051
LDL	2.88 ± 1.21	3.21 ± 1.01	2.55 ± 1.11	8.665	<0.001	0.061	0.074	<0.001c
	CSVD-CI (n = 61)	HC (n = 108)	CSVD-NC (n = 85)	CSVD-CI vs. HC vs. CSVD-NC		CSVD-CI vs. HC	CSVD-CI vs. CSVD-NC	HC vs. CSVD-NC
				F	p	p	p	p
LI	3.62 ± 1.45	0	2.71 ± 1.21	312.393	<0.001	<0.001b	<0.001a	<0.001c
WMH	8.23 ± 1.80	0	5.33 ± 1.77	810.436	<0.001	<0.001b	<0.001a	<0.001c
Smoking (1/0)	14/47	16/92	15/70	1.237	0.292	0.118	0.378	0.478
Alcohol consumption (1/0)	6/55	9/99	13/72	1.230	0.294	0.765	0.301	0.127
DM (1/0)	13/48	14/94	14/44	18.954	<0.001	0.213	<0.001a	<0.001c
HBP (1/0)	38/23	26/82	47/38	17.848	<0.001	<0.001b	0.272	<0.001c
CAD (1/0)	2/59	2/106	5/80	1.135	0.323	0.631	0.403	0.134
HL (1/0)	1/60	13/95	5/80	3.328	0.037	0.014b	0.105	0.334
MMSE score	22.57 ± 5.00	27.67 ± 1.64	27.21 ± 2.56	60.056	<0.001	<0.001b	<0.001a	0.305
HIS score	6.90 ± 3.21	0	6.45 ± 2.67	282.884	<0.001	<0.001b	0.22	<0.001c
HAMA score	4.30 ± 1.94	2.62 ± 2.09	3.87 ± 1.74	17.564	<0.001	<0.001b	0.195	<0.001c
HAMD score	3.07 ± 2.48	2.07 ± 2.05	2.78 ± 2.07	4.790	0.009	0.005b	0.428	0.026

Univariate analysis of variance and post hoc tests were used to assess differences in each group of continuous variables. Sex ratio differences were tested by the χ² test, and all data are presented as the mean ± standard deviation unless otherwise stated. Sex (1 male/0 female). Smoking (1 yes/0 no). Alcohol consumption (1 yes/0 no).

BMI, body mass index; BG, blood glucose; CAD, coronary heart disease (1 with/0 without); CSVD-CI, CSVD patients with cognitive impairment; CSVD-NC, CSVD patients with normal cognition; DBP, diastolic blood pressure; DM, diabetes mellitus (1 with/0 without); HAMA, Hamilton Anxiety Scale; HAMD, Hamilton Depression Scale; HBP, hypertension (1 with/0 without); HC, healthy control; HDL, high-density lipoprotein; HIS, Hachinski ischemia Scale; HL, hyperlipidemia (1 with/0 without); HR, heart rate; LDL, low-density lipoprotein; LI, lacunar infarction; MMSE, Mini-Mental State Examination; SBP, systolic blood pressure; TC, total cholesterol; WHH, white matter high signal.

^aSignificant difference between the CSVD-CI and CSVD-NC groups (p < 0.05).

^bSignificant difference between the CSVD-CI and HC groups (p < 0.05).

^cSignificant difference between the CSVD-NC and HC groups (p < 0.05).

Table 2 Results of neuropsychological tests in the CSVD-CI, HC, and CSVD-NC groups.

Cognitive indicators	CSVD-CI (n = 61)	HC (n = 108)	CSVD-NC (n = 85)	CSVD-CI vs. HC vs. CSVD-NC		CSVD-CI vs. HC	CSVD-CI vs. CSVD-NC	HC vs. CSVD-NC
				F	p	p	p	p
Attention
TMT_B_ACC	19.90 ± 5.86	24.20 ± 0.78	23.13 ± 2.77	33.316	<0.001	<0.001	<0.001	0.027
TMT_A_ACC	23.15 ± 1.89	24.06 ± 0.97	23.92 ± 1.11	10.406	<0.001	<0.001	<0.001	0.433
DST	7.39 ± 1.57	7.77 ± 1.19	7.94 ± 1.44	2.870	0.059	0.090	0.018	0.387
R_DST	3.18 ± 1.07	4.17 ± 0.94	3.66 ± 1.34	15.623	<0.001	<0.001	0.012	0.002
WM_Trial1	1.41 ± 1.00	3.16 ± 1.26	2.24 ± 1.13	46.209	<0.001	<0.001	<0.001	<0.001
WM_Trial2	1.26 ± 0.75	2.70 ± 1.52	1.92 ± 1.04	28.769	<0.001	<0.001	0.001	<0.001
WM_Total	2.38 ± 1.61	5.60 ± 2.19	4.04 ± 2.03	51.354	<0.001	<0.001	<0.001	<0.001
Memory
CFT_immediate_recall_score	29.15 ± 8.86	33.28 ± 3.73	32.36 ± 4.67	10.706	<0.001	<0.001	0.001	0.264
CFT_delayde_recall_score	8.45 ± 5.76	15.78 ± 6.73	12.86 ± 6.63	24.999	<0.001	<0.001	<0.001	0.002
RAVLT_immediately	4.37 ± 1.80	7.65 ± 1.55	6.99 ± 1.69	78.780	<0.001	<0.001	<0.001	0.007
RAVLT_delay_recall_5 min	4.15 ± 1.89	9.01 ± 2.10	7.44 ± 2.65	90.979	<0.001	<0.001	<0.001	<0.001
RAVLT_delay_recall_20 min	3.26 ± 1.92	8.56 ± 2.33	7.02 ± 2.62	100.661	<0.001	<0.001	<0.001	<0.001
Executive
S1_ACC	89.36 ± 15.50	106.80 ± 6.42	102.34 ± 8.46	60.823	<0.001	<0.001	<0.001	0.002
S2_ACC	78.07 ± 13.27	98.76 ± 10.83	91.25 ± 15.27	49.074	<0.001	<0.001	<0.001	<0.001
N_ACC	53.95 ± 30.44	54.32 ± 9.01	58.74 ± 21.61	1.428	0.242	0.909	0.161	0.135
Language
Boston	17.75 ± 5.11	22.56 ± 3.96	21.65 ± 3.96	33.077	<0.001	<0.001	<0.001	0.255
VFT (total)	25.41 ± 8.56	38.01 ± 4.94	33.57 ± 8.24	70.637	<0.001	<0.001	<0.001	<0.001
Visuospatial
CDT	2.61 ± 1.16	3.10 ± 0.93	3.20 ± 0.98	7.391	0.001	0.002	<0.001	0.486

Univariate analysis of variance and post hoc tests were used to assess differences in each group of continuous variables. Sex ratio differences were tested by the χ² test, and all data are presented as the mean ± standard deviation unless otherwise stated.

Boston, Boston Naming Test; CDT, clock drawing test; CFT, complex figure test; CSVD-CI, CSVD patients with cognitive impairment; CSVD-NC, CSVD patients with normal cognition; DST, digital span test; HC, healthy control; RAVLT, Rey's Auditory Verbal Learning Test; R_DST, reverse digit span test; S1_ACC, S2_ACC, N_ACC, Stroop test; VFT, verbal fluency task; WM, working memory.

Source-based morphometry processing

We found 13 meaningful independent components (ICs) using SBM. The thalamic covariance network, cerebellar covariance network (B), and calcarine covariance network ICs showed significant intergroup differences among the three groups (Fig. 1, Table 3). In addition, the expression of the thalamic covariance network significantly differed between CSVD-CI and CSVD-NC patients. The specific performance is as follows.

Figure 1. Abnormal gray matter covariance network in CSVD patients with and without cognitive impairment and healthy controls. (A) In IC1, the weight coefficient of the CSVD-CI group was lower than both that of the CSVD-NC group and that of the HC group. (B) In IC7, the weight coefficients of CSVD-CI and CSVD-NC groups were lower than those of the HC group. (C) In IC26, the weight coefficients of CSVD-CI and CSVD-NC groups were lower than those of the HC group.

Table 3 Analysis of gray covariance networks among the three groups.

Name of ICs	CSVD-CI	CSVD-NC	HC	ANOVA		HC vs. CSVD-CI		HC vs. CSVD-NC		CSVD-NC vs. CSVD-CI
				F	p	Mean difference	p corrected	Mean difference	p corrected	Mean difference	p corrected
Thalamus covariance network(IC1)	5.3 × 10−3 ± 2.5 × 10−4	6.4 × 10−3 ± 1.9 × 10−4	7.3 × 10−3 ± 2.2 × 10−4	18.479	<0.001	2.0 × 10−3 ± 3.3 × 10−4	<0.00033	9.0 × 10−4 ± 2.9 × 10−4	<0.00033	1.1 × 10−3 ± 3.2 × 10−4	0.00367
Cerebellar covariance network B(IC7)	1.3 × 10−2 ± 3.5 × 10−4	1.4 × 10–2 ± 2.7 × 10−4	1.5 × 10−2 ± 3.0 × 10−4	5.797	0.003	1.4 × 10−3 ± 4.6 × 10−4	0.00533	1.1 × 10−3 ± 4.1 × 10−4	<0.00033	–	–
Calcarine covariance network (IC26)	−1.5 × 10−3 ± 3.1 × 10−4	−6.0 × 10–4 ± 2.3 × 10−4	9.1 × 10−4 ± 2.6 × 10−4	18.992	<0.001	2.4 × 10−3 ± 3.6 × 10−4	<0.00033	1.5 × 10−3 ± 3.6 × 10−4	0.00033	–	–

The results are expressed as the mean ± standard deviation, and p < 0.05 was considered statistically significant. The significance level of the mean difference is p_corrected <0.05/K, K (indicates the number of comparisons) = 9, and pcorrected <0.0056. Multiple comparison correction: Bonferroni method. Correlation analysis of gray matter structural covariance networks with various cognitive indicators in patients with CSVD.

IC1 included the thalamus and parahippocampal gyrus. In IC1, the weight coefficient of the CSVD-CI group (5.3 × 10⁻³ ± 2.5 × 10⁻⁴) was lower than that of the CSVD-NC group (6.4 × 10⁻³ ± 1.9 × 10⁻⁴), and both groups showed lower weight coefficients than the HC group (7.3 × 10–3 ± 2.2 × 10⁻⁴). IC7 included the cerebellum. In IC7, the weight coefficients of the CSVD-CI group (1.3 × 10⁻² ± 3.5 × 10⁻⁴) and CSVD-NC group (1.4 × 10⁻² ± 2.7 × 10⁻⁴) were lower than those of the HC group (1.5 × 10⁻² ± 3.0 × 10⁻⁴). IC26 included the calcarine sulcus and precuneus. In IC26, the weight coefficients of the CSVD-CI group (−1.5 × 10⁻³ ± 3.1 × 10⁻⁴) and CSVD-NC group (−6.0 × 10⁻⁴ ± 2.3 × 10⁻⁴) were lower than those of the HC group (9.1 × 10⁻⁴ ± 2.6 × 10⁻⁴). The specific anatomical locations of these gray covariance networks are shown in Table 4.

Table 4 Anatomical labels of covariance networks of gray matter structures showing significant differences between the three groups.

Anatomical regions	Brodmann area	Volume (cc) left/right	Maximum z value for left/right hemisphere (MNI coordinates x, y, z)
IC1
Thalamus	–	7.6/6.9	17.8 (−11, −15, 7)/18.0 (11, −13, 7)
Hippocampus	27	0.1/0.0	3.7 (−19, −35, 2)/na
IC7
Cerebellar_Crus1	–	20.1/20.5	8.7 (−31, −79, −24)/8.4 (30, −81, −31)
Cerebellar_6	–	2.0/1.5	6.7 (−30, −77, −21)/6.7 (21, −88, −26)
Cerebellar_Crus2	–	1.9/3.7	6.2 (−49, −64, −40)/6.3 (23, −74, −37)
Fusiform Gyrus	19	0.1/0.0	3.8 (−40, −69, −17)/na
Vermis_8	–	0.2/0.4	3.1 (−4, −65, −32)/3.6 (−1, −69, −31)
IC26
Calcarine	17, 18, 19, 23, 37	12.1/11.9	13.2 (−23, −64, 6)/11.5 (22, −60, 5)
Lingual Gyrus	19	0.9/1.3	7.9 (−22, −53, 1)/9.8 (24, −52, 1)
Precuneus	30, 23	0.8/0.2	4.5 (−3, −50, 26)/4.0 (1, −49, 30)
Precentral Gyrus	6	0.4/0.0	3.8 (−43, −7, 50)/na
Inferior Frontal Gyrus	45	0.1/0.0	3.1 (−40, 40, 8)/na

The correlations between cognitive indicators and the weight coefficient of the gray matter structure covariance network in patients with CSVD

As shown in Figure 2, in CSVD patients, the weight coefficient of IC1, which included the thalamus and parahippocampal gyrus, was positively correlated with attention (r = 0.199, p = 0.047) (A), language (r = 0.172, p = 0.047) (B), memory (r = 0.193, p = 0.047) (C), and visual–spatial functions (r = 0.176, p = 0.047) (D).

Figure 2. Partial correlation results of cognitive deficit scores and gray matter covariance network weight coefficients of CSVD patients showing that attention, language, memory, and visuospatial function were positively correlated with IC1 weight coefficients (A–D).

Support vector machine analysis results

As shown in Figure 3, by using the weight coefficient of the thalamic covariance network, which significantly differed between CSVD-CI and CSVD-NC patients, the constructed prediction model had a significant classification effect and good performance, with an AUC of 0.86, accuracy of 0.80, specificity of 0.77, and sensitivity of 0.88.

Figure 3. ROC curves of the composite prediction model constructed using the weight coefficients of thalamic covariance networks that showed intergroup differences between CSVD-CI and CSVD-NC patients combined with the cognitive score scale (excluding the measures used to group CSVD patients based on cognitive function). AUC, area under the curve.

Results of the analysis after the CSVD patients with thalamic lacunes were excluded

In this study, some of our included patients in CSVD may have had thalamic lacunes. Thalamic lacunes may disrupt thalamic covariance network expression, so we screened the MRI data of the included CSVD patients. According to the above criteria, we excluded a total of 29 CSVD patients with thalamic lacunes (including 17 CSVD-CI patients and 12 CSVD-NC patients) and repeated the above steps for the remaining 225 subjects (including 44 CSVD-CI patients, 108 healthy volunteers, and 73 CSVD-NC patients). The detailed results are shown in Supplementary Material S2.

Discussion

In the present investigation, SBM analyses of subjects included and excluded thalamic lacunes revealed significant decreases in particular gray matter structure covariance networks among CSVD patients, both with and without CI, in contrast to HCs. Remarkably, the thalamic covariance network, involving the thalamus and parahippocampal gyrus, demonstrated substantially lower coefficient weights in the CSVD-CI group than in the CSVD-NC group. These findings suggest that the weight coefficient of the thalamic covariance network is associated with cognitive performance in individuals with CSVD. Consequently, our results point toward the potential role of the weight coefficient of the thalamic covariance network as a distinguishing feature for identifying CI patients within the CSVD cohort.

The results of the analysis of all subjects indicated CSVD patients demonstrated notably reduced coefficient weights in the thalamic covariance network, cerebellar covariance network, and calcarine covariance network compared to the HCs. This finding is consistent with a prior study that reported reductions in these regions among CSVD-CI patients.¹⁷ Both studies observed reductions in structural covariance networks involving the cerebellum, temporal lobe, cingulate gyrus, parahippocampal gyrus, precuneus, frontal lobe, and precentral gyrus in CSVD patients. In addition, Foster-Dingley et al.¹⁷ detected decreases in structural covariance networks involving the amygdala, a region involved in the regulation of mood, behavior, and memory,²³ the insular cortex, which is associated with the regulation of perception and language function,²⁴ and the hippocampus, which participates in modulating memory and perception²⁵ in CSVD-CI. The discrepancies in the findings may be attributed to three factors: (i) Foster-Dingley et al.¹⁷ included subjects aged over 75 years with CI, whereas the current study included subjects aged between 50 and 75 years presenting with and without CI. Structural covariance networks may begin to alter or become detectable with increased age or vary based on disease severity. (ii) The former study defined cognitive impairment simply as a Mini-Mental State Examination (MMSE) score <27, whereas we accounted for age and years of education, employing cognitive unit scores to categorize subjects¹⁹; (iii) finally, in the earlier study, data were self-clustered into eight spatial components based on experience, whereas our study utilized a data-driven approach, namely, the number of components was determined through automatic identification.¹⁰

Our research expands upon prior MRI studies exploring the brain structure of CSVD patients. In our investigation, a subgroup analysis of CSVD patients (including CI and NC patients) revealed that patients with CSVD-CI had significantly reduced coefficient weights in the thalamic covariance network, encompassing the thalamus and parahippocampal gyrus, compared to those with CSVD-NC. Both the thalamus and the parahippocampal gyrus represent core regions for early atrophy in the brains of CSVD patients, contributing to atrophy in other brain areas and are associated with cognitive function.²⁶ Our study may offer a neurobiological marker for the early detection of CSVD-CI and CSVD-NC patients.

Our research identified a positive correlation between the thalamic covariance network and attention, language, memory, and visuospatial functions. This network encompasses the thalamus and the parahippocampal gyrus, and our findings are consistent with prior studies that demonstrated an association between lesions in the parahippocampal gyrus^17,27 and memory function. Past research has indicated that the thalamus plays a pivotal role in neural signal transduction and cognitive function modulation, with the thalamocortical circuit being critical for perception–cognition conversion.^28–31 A study by Li et al. found no significant correlation between cognitive deficit and thalamic volume redu ction in CSVD patients, but a significant negative correlation was observed with thalamic-subcortical white matter fiber bundle connections.³² Our research not only complements but also elucidates the findings of prior studies examining the correlation between cognitive function and structural covariance networks in CSVD patients. Consistent with the findings of Foster-Dingley et al.¹⁷, we did not find a correlation between any structural covariance networks and executive function. Previous research on 19-79-year-old healthy individuals revealed a positive correlation between the covariance network of gray matter structures, including the prefrontal cortex and anterior cingulate cortex, and executive function.³³ However, this is contrary to our findings, potentially due to differences in age and health status of the subjects in our study.

Previous studies have shown that white matter hypersignal can potentially disrupt the distal white matter fiber tract network, which may lead to cognitive impairment in patients with CSVD.^34–36 Our study supports and enriches previous studies. Our results indicate that thalamic covariance network expression remained significantly lower in CSVD-CI patients after excluding patients with thalamic lacunes than in CSVD-NC patients. Our findings provide new evidence to explain the distribution status of whole-brain gray matter reduction in CSVD patients, and the thalamic covariance network is the core region for early gray matter reduction during the development of CSVD disease.

Moreover, in the analysis of patients excluding thalamic lacunes, we identified a positive relationship between the thalamic covariance network and attention, while no significant association was found with the other cognitive units. The inconsistency with previous studies may be due to the presence of sample size or the potential impact of thalamic lacunes.

The classification model we developed, which employs the weight coefficients of the thalamic covariance network in combination with the cognitive score scale data of CSVD patients (excluding those used to determine CI vs. NC grouping), demonstrated high accuracy and replicability. A comprehensive analysis integrating the gray matter structural network and cognitive scale information could provide a more objective method for assessing the onset of cognitive impairment in CSVD patients.

Previous research has indicated that different brain morphological indicators have various advantages.²¹ Hence, integrating multiple morphological indicators can offer a more comprehensive depiction of brain morphology in CSVD patients. Future studies should consider combining several morphological indicators to deepen our understanding of brain morphological alterations linked to cognitive deficits in CSVD patients.

The present study is not without its limitations. Primarily, given its cross-sectional design, we could not ascertain the temporal dynamics of cognitive deficits and structural covariance network anomalies in CSVD patients. Therefore, future research should employ a longitudinal approach to track changes in the gray matter structure covariance network over time in CSVD patients. Scholars should also investigate the correlation between dynamic changes in this network and cognitive function. Moreover, the association between the gray matter structure covariance network and responses to clinical treatment merits further exploration. Additionally, the differences in the structural covariance network between the CSVD-CI and CSVD-NC groups, as identified via the SBM approach, may not fully capture individual-level voxel changes due to heterogeneity in factors such as age, brain volume, and CSVD severity. Consequently, the findings of this study do not allow for population-level inferences about cognitive impairment in CSVD individuals. Last, in this study, it could not be determined whether thalamic covariance network reduced expression in the CSVD-CI group relative to the NC group was really caused by the altered microstructural properties of brain tissue (including iron and water content both inside or outside of neuronal cells, formation of myelination, etc.) in the CSVD, which is a potential confounding factor that future studies must face.

Conclusion

Our results suggest that the gray matter structural covariance network could explain the irregular distribution of gray matter reduction in patients with CSVD. Our findings reveal a significant reduction in the thalamic covariance network, which encompasses the thalamus and the parahippocampal gyrus, in CSVD-CI patients relative to NC patients. Furthermore, the coefficient weights of this network correlate with the cognitive deficits seen in CSVD patients. The classification model we constructed, which incorporates the weight coefficients of the thalamic covariance network and the cognitive score scale, can effectively identify CSVD patients with cognitive impairment at an early stage.

Author Contributions

Wei Yan: Conceptualization, methodology, data curation, formal analysis, visualization, writing—review & editing. Siwei Tang and Ting Lei: Visualization, writing—review & editing. Haiqing Li, Li Chen, Miao He, Lijing Zhou: Writing—review & editing. Yajun Li, Yuxing Jiang, Xi He: Conceptualization, data curation, writing—review & editing. Hongjian Li and Chen Zeng: Conceptualization, data curation, writing—review & editing.

Acknowledgements

We thank Sichuan Science and Technology Program (2019YJ0380), Nanchong Science and Technology Program (20SXZRKX0011) and National Clinical Key Specialty Construction Sichuan Provincial Health Commission for sponsoring this study. Thanks to the laboratory members for the acquisition and processing of the data for this study. Thanks to the ATCN editor's approval and the review teacher's guidance.

Conflict of Interest

The authors report no conflicts of interest.