2.1 Participants

This was a case–control study involving 121 right-handed participants. The protocol of this study was approved by the Medical Ethics Committee of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, Shanghai University of Traditional Chinese Medicine. At baseline, all participants underwent a neuropsychological evaluation and MRI conducted by professional imaging staff. They were also provided with information about the study and were required to sign informed permission forms.

The trail inclusion criteria were as follows: (i) 50 < age < 75; (ii) Summary Mental State Examination (MMSE) score ≥ 24; (iii) Hamilton Depression Scale (HAMD) score scale < 8 points; (iv) Modified Rankin Scale (MRS) score of ≤ 1; and (v) Hamilton Anxiety Scale (HAMA) rating scale < 7 points; (vi) right-handedness; (vii) National Institutes of Health Stroke Scale (NIHSS) score < 2. For VCI participants, they must meet the diagnostic criteria as defined by consensus guidelines [21], provide objective evidence of cognitive impairment or decline, characterized by at least 1.0 standard deviation below age- and education-adjusted norms in any cognitive domain, and have white matter lesions, as evidenced by Fazekas scores of 1 or greater. HCs must show no signs of cognitive impairment in any cognitive domain and must not have any white matter lesions, as indicated by a Fazekas score of 0 [22]. Exclusion criteria were as follows: (i) MRI indicating other brain parenchymal lesions (e.g., malformation or tumor); (ii) history of other neurological or psychiatric disorders (e.g., Alzheimer's disease, epilepsy, Parkinson's disease, ans schizophrenia); (iii) severe cardiac, hepatic, or renal insufficiency; (iv) inability to complete neuropsychological assessments; (v) inability to undergo brain MRI.

2.2 Neuropsychological Assessments

Neuropsychological assessments were conducted by a neuropsychologist who was blinded to the clinical characteristics of the participants. A second neuropsychologist, also blinded to the clinical details, performed the scoring of the assessments. The neuropsychological assessment included the Mini-Mental Status Examination (MMSE) scores [23] and performance in four cognition domains: (1) episodic memory: Auditory Verbal Learning test (AVLT) [24, 25], (2) working memory: Trail Making test (TMT) (B time) [26], (3) attention: Symbol Digit Modalities test (SDMT), [27] and (4) language tests: Boston Naming Test (BNT) [28].

2.3 Fazekas Score Evaluation

The WMH grade was assessed using the Fazekas classification method on T2-weighted fluid-attenuated inversion recovery imaging (FLAIR). PVWMH and DWMH were scored separately and summed for the total Fazekas score. PVWMH scores: 0 (none), 1 (caps/pencil-thin lining), 2 (smooth halo), and 3 (irregular). DWMH scores: 0 (none), 1 (punctate), 2 (beginning confluence), and 3 (large confluent). Total scores: mild (1, 2), moderate (3, 4), and severe (5, 6). This procedure was evaluated by two neurologists who were unaware of the participants' clinical features. In cases of discrepancies in MRI interpretations, a third neurologist reviewed the images to reach a consensus. This double-blind methodology ensured the objectivity and reliability of the neuropsychological evaluations and imaging analyses.

2.4 MRI Acquisition

MRI data were acquired using a Siemens MAGNETOM Verio, a high-field 3.0 Tesla scanner manufactured in Erlangen, Germany. Participants were guided to keep their eyes shut, remain tranquil, and avoid dozing off throughout the MRI scan. Resting-state functional MRI images were captured through a gradient-echo echo-planar imaging sequence with the following specifications: axial orientation; repetition interval of 3000 ms; field of view measuring 203 mm × 203 mm; matrix dimensions of 64 × 64; flip angle set at 52°; slice thickness of 3.0 mm; 43 slices; no gap (voxel size equates 3.6 mm × 3.6 mm by 3.0 mm); and a total of 200 volumes acquired. The diffusion-weighted imaging dataset was collected with these parameters: TR at 10,000 ms, TE at 89 ms, flip angle at 90°, matrix size of 128 × 128, FOV of 240 × 240 mm, slice thickness of 2.0 mm, and resolution of 1.875 × 1.875 × 2.0 mm. The DWI series includes 60 diffusion-weighted images taken across 60 non-collinear gradient directions with a b-value of 1000 s/mm², complemented by two additional images with a b-value of 0 s/mm². High-resolution T1-weighted structural images were acquired using a 3D magnetization-prepared rapid gradient-echo (T1WI-3D-MPRAGE) sequence. The scanning parameters were as follows: TR = 1900 ms, TE = 2.93 ms, flip angle = 9°, matrix = 256 × 256, FOV = 256 mm × 256 mm, slice thickness = 1.0 mm, and resolution = 1 × 1 × 1 mm.

2.5 Preprocessing

Details of MRI data preprocessing can be found in the Supporting Informations.

2.6 Spatial Statistics Along White Matter Tracts

A voxel-based analysis using TBSS was applied to evaluate fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) metrics. The methodology consisted of several steps: (i) registration of individual FA images to the MNI standard space using linear and nonlinear techniques [29]; (ii) creation of a mean FA template and a fiber skeleton thresholded at 0.2 [29]; (iii) projection of aligned FA images onto the mean FA skeleton for voxel-wise analysis; (iv) application of TBSS to MD, AD, and RD data using the FSL software suite [30]; and (v) extraction of mean FA, MD, AD, and RD values for each tract in the Johns Hopkins University (JHU) White Matter Tractography Atlas [31] for group comparisons.

2.7 Structural Network Construction and Graph Theoretical Analysis

The individual FA skeletons constructed as described above were used to build structural brain networks. The structural connectivity network was modeled as an unweighted network comprising 90 nodes, defined by the automated anatomic labeling (AAL) template [32]. The fiber assignment by continuous tracking (FACT) algorithm was employed to track white matter fibers between node pairs [33]. Fiber tracking was terminated when the FA value dipped below 0.2 or the tracking angle surpassed 45° [29, 34]. The connectivity within the brain network was quantified by the count of fibers linking each pair of regions, culminating in a 90 × 90 network matrix. The binary threshold of retaining connections in over 80% of participants was chosen to ensure reliable connections. By focusing on connections present in the majority, we reduce the impact of individual variability and noise, highlighting more stable and biologically meaningful connections [35]. If the fiber count between nodes $$$ i $$$ and $$$ j $$$ was 3 or more, a binary value of 1 was designated for the connection $$$ \left(i,j\right) $$$ element $$$ \left({e}_{j,j}\right) $$$. Conversely, if the count was less than 3, $$$ {e}_{j,j} $$$ was set to 0. This thresholding method significantly minimizes the risk of spurious connections due to noise or limitations inherent in deterministic tractography [36]. Consequently, a binary 90 × 90 matrix was assembled, representing the structural connectivity of the white matter in the brain. Subsequent binary graph theory-based network analyses were performed in the GRETNA toolbox [37]. Several key graph measures were calculated, including global efficiency (Eglob), local efficiency (Eloc), shortest path length (Lp), clustering coefficient (Cp), small-world parameters (Sigma), nodal efficiency (NEglob), nodal local efficiency (NEloc), betweenness centrality, degree centrality, nodal clustering coefficient (NCp), and nodal shortest path length (NLp). The definition and calculation formula can be found in the Supporting Informations.

2.8 Functional Network Construction and Graph Theoretical Analysis

Following the de-noising process, the functional connectivity between each pair of brain regions was measured by computing the Pearson correlation coefficient between the average BOLD time series of the regions. A 90 × 90 weighted adjacency matrix, representing the connectome, was then created for each participant. Fisher Z-transformation was used to increase the normal distribution properties of the network. The sparsity threshold range was set from 0.01 to 0.40 with a step size of 0.01 to generate binary undirected networks characterized by small-world architecture and connectivity. Based on the constructed functional network matrix, GRETNA was also used to calculate the topological properties of the functional network, including global efficiency (Eglob), local efficiency (Eloc) shortest path length (Lp), clustering coefficient (Cp), small-world parameters (Sigma), nodal efficiency (NEglob), nodal local efficiency (NEloc), betweenness centrality, degree centrality, nodal clustering coefficient (NCp), and nodal shortest path length (NLp).

2.9 Structure–Function Coupling Analysis

The regional connectivity signatures were derived from individual columns of each participant's structural or functional connectivity matrix, encapsulated as vectors that quantify the strength of connections emanating from a single network node to the remainder of the nodes. The coupling between the structure and function was quantified using the Spearman rank correlation applied to the non-zero components of both structural and functional connectivity profiles [38, 39]. These regional measures of structure–function coupling were subsequently averaged across the cohort of participants.

2.10 Numerical Data Statistical Analysis

Data were analyzed using SPSS 25.0 software (IBM, Armonk, NY, USA). The Shapiro–Wilk test was employed to assess the normality of numerical data; if the p value from the test was greater than 0.05, the data were considered normally distributed. For MRI image data, Fisher's Z-transformation was applied to enhance the normality of the data. Normally distributed continuous variables were analyzed with the independent t-test, with results presented as mean ± standard deviation. For non-normally distributed continuous variables, the Mann–Whitney U test was used, and the results were expressed as median and interquartile range. Categorical variables were expressed as percentages and compared using the chi-squared test. Pearson's correlation analysis was applied for normally distributed data, while Spearman's correlation analysis was used for non-normally distributed or categorical data. In the group comparison process, linear regression was employed to control for the effects of age, gender, and education level on statistical outcomes. Following this regression of covariates, partial correlation analysis was conducted to explore the relationship between brain network attributes and cognitive performance. All hypothesis tests were two-tailed, and p values < 0.05 were considered significant. Family-wise error (FWE) correction was applied to reduce errors from multiple comparisons.

3.1 Participants

A total of 130 subjects were recruited into the study at first. All participants underwent neuropsychological assessment and MRI scanning, including DTI and BOLD images. We visually inspect the MRI images of the subjects and exclude those with excessive artifacts. One subject with severe MRI imaging artifacts and two subjects with MMSE scores below 24 were excluded after MRI evaluation. The images of participants whose head movements exceeded 3.0 mm or whose rotations were over 3.0° were considered to have an impact on the analysis results and thus were excluded. Following image preprocessing, six subjects with excessive head movements were further excluded, resulting in 121 subjects included in the final analysis. Subjects were categorized into HCs (n = 53) versus VCI participants (n = 68) based on cognitive scales and Fazekas scores. Finally, the data were analyzed statistically (Figure 1).

Participants' baseline demographic characteristics were comparable between the two trial groups. The VCI group exhibited inferior cognitive performance across memory, executive function, information processing speed, attention, and naming ability. Detailed information is available in Table 1.

3.2 Variability of Diffusion Tensor Metrics in VCI Individuals

In comparison to the HCs, VCI participants exhibited significant damage to their whole-brain white matter fiber tracts. To quantify these changes, FA, MD, AD, and RD values were extracted and subjected to statistical analysis between the groups. The results revealed that, compared to HCs, FA values were significantly decreased in VCI participants. Conversely, MD and RD values were significantly increased in VCI participants, indicating compromised white matter integrity (Figure 2).

3.3 Structural Network Connectivity

In analysis of structural network topological properties, participants with VCI showed lower global structural connectivity compared to HCs, specifically with reduced global efficiency (t = 2.789, p = 0.006), lower local efficiency (t = 2.528, p = 0.013), diminished small-world properties (t = 2.642, p = 0.009), and increased characteristic path length (t = −2.693, p = 0.008) (Figure 3A).

As for nodal properties, VCI participants exhibited statistically significant decreases in the degree centrality of the left posterior cingulate gyrus, right precuneus, right posterior cingulate gyrus, right middle occipital gyrus, right cuneus, right superior parietal gyrus, right postcentral gyrus, right putamen, and right amygdala, as well as the left putamen, compared to HCs (p < 0.05, FWE-corrected). Participants with VCI had a decreased nodal efficiency in left posterior cingulate gyrus (PCG.L), right precuneus (PCUN.R), left precuneus (PCUN.L), right posterior cingulate gyrus (PCG.R), right middle occipital gyrus (MOG.R), right cuneus (CUN.R), right calcarine cortex (CAL.R), right superior parietal gyrus (SPG.R), right postcentral gyrus (PoCG.R), and left precentral gyrus (PreCG.L) (p < 0.05, FWE-corrected). We also found increased nodal shortest path in left posterior cingulate gyrus (PCG.L), right precuneus (PCUN.R), left precuneus (PCUN.L), right middle occipital gyrus (MOG.R), right cuneus (CUN.R), right calcarine cortex (CAL.R), and left precentral gyrus (PreCG.L) in VCI participants compared to that in HCs (p < 0.05, FWE-corrected) (Figure 3B). No significant differences in betweenness centrality, nodal local efficiency, or nodal clustering coefficient were found between participants with VCI and HCs.

3.4 Functional Network Connectivity

The analysis of functional network topological properties showed no statistically significant differences between VCI participants and HCs in global efficiency, local efficiency, clustering coefficient, characteristic path length, and small-worldness (Sigma) (Figure 4). Additionally, intergroup comparisons revealed no nodal property differences after FWE correction.

3.5 Structure–Function Coupling

A positive correlation trend between the FA structural and resting-state functional matrices of participant HC 001 was unfolded into vectors (Figure 5A). This trend was also evident in participant HC 001's network graph, where regions of interest (ROIs) with structural connectivity showed strong functional connectivity (Figure 5B).

Group comparisons of nodal coupling relationships revealed that VCI participants had higher structure–function coupling in bilateral calcarine cortex, bilateral cuneus, bilateral superior occipital gyrus, bilateral lingual gyrus, bilateral insula, bilateral middle temporal gyrus, bilateral postcentral gyrus, bilateral inferior temporal gyrus, bilateral fusiform gyrus, right middle frontal gyrus, left inferior occipital gyrus, left superior orbital gyrus, left hippocampus, right superior temporal gyrus, right dorsolateral superior frontal gyrus, left supplementary motor area, right paracentral lobule, right precuneus, left precentral gyrus, and left Rolandic operculum compared to HCs (p < 0.0001, FWE-corrected) (Figure 5C).

3.6 The Impact of Structure–Function Coupling in Cognitive Function

To investigate the increased structural–functional coupling in VCI patients compared to HCs, partial correlation analyses were conducted to assess the relationship between structural–functional coupling in differing regions and performance on the BNT, STT_B, SDMT, and AVLT (Figure 6). The results showed a significant negative correlation between coupling in the Calcarine_R, Lingual_R, Occipital_Inf_L, Fusiform_L, Temporal_Sup_R, and Temporal_Mid_R regions and AVLT scores, which reflects memory function (p < 0.001, FWE-corrected). Additionally, a significant negative correlation was observed between coupling in the Insula_R, Calcarine_R, Lingual_R, Occipital_Sup_L, Occipital_Inf_L, Fusiform_L, and Temporal_Sup_R regions and SDMT scores, reflecting information processing speed (p < 0.001, FWE-corrected). The uncorrected outcomes are detailed in the Supporting Informations.

4.1 Limitations

This study has several limitations. Despite employing multiple neuroimaging techniques, the cross-sectional design prevents establishing causality between white matter integrity and cognitive impairment in VCI. Additionally, the reliance on cognitive scales and Fazekas scores for classification might not comprehensively reflect the spectrum of cognitive decline or white matter changes. Potential head motion during MRI scanning may compromise the accuracy of DTI metrics. Moreover, the lack of significant differences in functional network connectivity analysis might indicate limitations in methodological sensitivity. Future research should include larger, longitudinal studies to validate and extend these findings.

4.2 Conclusions

This study identified notable neurocognitive and brain connectivity differences between individuals with VCI and HCs. VCI participants exhibited poorer cognitive performance and significant disruptions in white matter integrity, evidenced by decreased FA and increased MD and RD. Structural network analysis indicated reduced global and local efficiency, diminished small-world properties, along with altered nodal connectivity in several brain regions. Additionally, the differences in structural–functional coupling in these brain regions were negatively correlated with memory function and information processing speed. These findings highlight the extensive neural alterations in VCI and underscore the importance of integrating structural and functional connectivity analyses to understand the mechanisms of cognitive impairment.