Brain structural volume changes and gait disorders in patients with cerebral small vessel disease: a principal component and logistic regression analysis

Abstract

Aims

To investigate the relationship between the relative volume changes of brain structures and gait disorders in patients with cerebral small vessel disease (CSVD), and to identify key brain volume indicators that influence gait disorders (GD) through principal component analysis (PCA) combined with logistic regression analysis.

Methods

A total of 90 patients with CSVD were included and divided into two groups based on their Timed Up and Go Test (TUG) results: the CSVD gait disorder group (GD group, N = 50, TUG ≥ 11.5 s) and the CSVD non-gait disorder group (NGD group, N = 40, TUG < 11.5 s). Brain structural volumes and associated imaging markers were calculated using a 3.0T superconducting magnetic resonance imaging (MRI) system combined with an intelligent image analysis platform. The relative volume data of brain structures were subjected to PCA for dimensionality reduction. Regression analysis of the principal components was performed to identify key predictive indicators for gait disorders.

Results

Principal components 2 and 3 exhibited significant predictive power for gait disorders. Component 2 primarily represented cerebrospinal fluid, the temporal lobe, and the frontal lobe, while Component 3 primarily represented the third ventricle, lateral ventricle, and corpus callosum. Analysis of component loadings indicated that the expansion of cerebrospinal fluid-related structures was positively correlated with the severity of gait disorders, while atrophy of brain tissue structures showed a negative correlation. No significant difference was observed in the total CSVD burden score between the GD and NGD groups.

Conclusion

The enlargement of the cerebrospinal fluid spaces and the ventricular system are key factors contributing to gait disorders in CSVD patients. Additionally, atrophy in cortical areas such as the temporal and frontal lobes, as well as the corpus callosum, plays a predictive role in the severity of gait disturbances. The total CSVD burden score does not adequately reflect the degree of gait impairment and requires further refinement for clinical application.

Introduction

Cerebral small vessel disease (CSVD) represents one of the most common vascular pathologies of the aging brain and is a major contributor to mobility decline and gait disorders in older adults [1]. Clinically, gait impairment in CSVD is associated with reduced walking speed, impaired balance, increased fall risk, and subsequent loss of independence, thereby substantially affecting quality of life and long-term prognosis [2–4]. The mechanisms underlying gait disturbance in CSVD are complex and multifactorial. Neuroimaging studies have traditionally focused on white matter hyperintensities, lacunar infarctions, and global brain atrophy as key structural substrates. However, growing evidence suggests that gait control relies on distributed neural networks involving cortical regions, subcortical nuclei, interhemispheric connections, and cerebrospinal fluid dynamics, rather than on isolated lesions alone. Structural alterations affecting these interconnected systems may therefore exert cumulative effects on gait performance that are not fully captured by conventional single-marker assessments [5]. In routine clinical practice, commonly used imaging markers such as the Fazekas score, medial temporal lobe atrophy (MTA), Evans index, and total CSVD burden score provide useful summaries of disease severity. Nevertheless, these markers primarily reflect localized or aggregated pathological features and may lack sensitivity to multiregional structural patterns relevant to complex motor functions such as gait. In particular, the total CSVD burden score has shown inconsistent associations with functional outcomes, including gait impairment, across different studies. Recent advances in intelligent imaging analysis and automated brain segmentation have enabled precise quantification of regional brain volumes across multiple structures simultaneously [6, 7]. These developments offer new opportunities to explore how coordinated changes in cortical, subcortical, and ventricular structures relate to gait dysfunction. Multivariate approaches that integrate information across regions may therefore provide more comprehensive insights than traditional univariate analyses. In this context, principal component analysis (PCA) provides a data-driven framework to identify dominant patterns of covarying brain structural changes, while reducing dimensionality and multicollinearity among imaging variables. When combined with regression modeling, PCA may help to clarify which structural volume patterns are most strongly associated with gait disturbance in CSVD patients.

Accordingly, the present study applies PCA and logistic regression to relative brain volume measures derived from automated MRI analysis in patients with CSVD. The aims were to (1) characterize multiregional structural volume patterns associated with gait disturbance, (2) evaluate their relationship with established clinical imaging markers, and (3) assess whether PCA-derived patterns provide incremental value for understanding gait impairment beyond conventional CSVD metrics.

Data and methods

Research subjects

The study included 90 patients diagnosed with CSVD, who were admitted to the First Affiliated Hospital of Dalian Medical University between February 2023 and October 2024. In addition to age, the groups were matched based on demographic characteristics (e.g., gender, years of education) and clinical risk factors (e.g., smoking, hypertension, diabetes, coronary heart disease). This study was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Dalian Medical University. Prior to enrollment, all participants were provided with detailed information about the study and gave written informed consent.

Enrollment criteria

① Diagnostic criteria: Meets the latest diagnostic standards for CSVD [8], including recent imaging findings such as small subcortical infarctions, lacunar cerebral infarctions, cerebral hemorrhage, subcortical white matter lesions, cortical iron deposition, cerebral microbleeds (CMB), and microinfarctions.

② Clinical manifestations: Presents with CSVD-related clinical symptoms, including cognitive impairment, gait disturbances, emotional disorders, and urinary or bowel dysfunction, in conjunction with relevant risk factors.

③ Age range: Aged between 50 and 80 years.

④ Ability to participate: Capable of cooperating with the gait analysis task.

⑤ Functional status: Modified Rankin Scale (mRS) score ≤ 3.

⑥ Informed consent: Voluntarily participates in the study and provides written informed consent.

Exclusion criteria

① Acute stroke: Patients in the acute stage of stroke.

② Neurological disorders: Individuals with neurological conditions such as epilepsy, multiple sclerosis, Parkinson’s disease, or Alzheimer’s disease.

③ Musculoskeletal disorders: Patients with spinal diseases, lower limb bone diseases, or limb disabilities that impair the ability to cooperate with gait tasks.

④ History of significant cerebral events: A history of traumatic brain injury, brain surgery, cerebral hemorrhage, or significant stenosis (> 50%) of head and neck arteries, as well as severe comorbid medical conditions.

⑤ Hydrocephalus: Any form of hydrocephalus, regardless of etiology.

⑥ Special subtypes of CSVD: Patients with specific types of CSVD, including cerebral amyloid angiopathy, hereditary small vessel disease, or immune-mediated small vessel disease.

⑦ MRI contraindications: Patients unable to undergo MRI due to contraindications.

Grouping method

Based on the results of the TUGtest, patients with a TUG time ≥ 11.5 s were assigned to the CSVD gait disturbance group (GD group), comprising 50 participants, while those with a TUG time < 11.5 s were assigned to the CSVD non-gait disturbance group (NGD group), comprising 40 participants.

Research methods

This study is both a cross-sectional and a case-control study. By employing these two methodologies, it aims to comprehensively analyze the imaging differences associated with gait disorders in patients with CSVD, elucidate their underlying pathophysiological mechanisms, and provide a scientific foundation for optimizing clinical diagnostic and treatment strategies. (Refer to Fig. 1 for the flow chart)

Fig. 1.

the flow chart represents the study participants sampling method

Imaging examination

All subjects underwent brain scans using a standardized 3.0T superconducting MRIscanner, with a scanning protocol that included T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), fluid-attenuated inversion recovery (FLAIR), diffusion-weighted imaging (DWI), and susceptibility-weighted imaging (SWI). During the scanning procedure, the patient’s head was immobilized within the scanner to minimize the impact of motion artifacts.

Gait assessments

A 3-meter TUG test was utilized in this study. The TUG test is a widely used, rapid, quantitative assessment tool for evaluating balance and walking function. It is primarily employed to assess the mobility of elderly individuals or patients and to predict their risk of falls [9]. In this study, the subjects’ balance, walking speed, turning ability, and lower limb muscle strength were comprehensively evaluated by measuring the time taken to rise from a chair, walk 3 m, turn around, and sit down again. The cutoff value of 11.5 s for the TUG test was selected based on prior studies demonstrating that this threshold effectively discriminates individuals with impaired mobility and increased fall risk in older adults and neurological populations. Previous work has shown that TUG values ≥ 11–12 s are associated with functional gait impairment and reduced balance performance, particularly in populations with cerebrovascular and neurodegenerative conditions. Given the age range and clinical characteristics of the CSVD cohort in the present study, a cutoff of 11.5 s was considered clinically appropriate and consistent with existing literature.

Artificial intelligence medical imaging analysis

Brain structural segmentation and quantitative imaging analysis were performed using uAIDiscoverCSVD (Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China; software version 1.3.0), an artificial intelligence-based medical imaging analysis platform specifically developed for CSVD. The system is built on deep learning algorithms trained on large-scale, manually annotated brain MRI datasets and is designed to automatically identify, segment, and quantify CSVD-related brain structures.

The uAIDiscoverCSVD system provides automated calculation of commonly used clinical imaging markers, including the total CSVD burden score, medial temporal lobe atrophy (MTA) score, Fazekas score, and Evans index. In parallel, it performs volumetric quantification of a wide range of brain structures by extracting region-specific boundary information from MRI images. Identifiable structures include, but are not limited to, total brain volume, gray matter, white matter, cerebrospinal fluid (CSF), peripheral CSF spaces, third ventricle, fourth ventricle, lateral ventricles, hippocampus, globus pallidus, thalamus, insular lobe, temporal lobe, frontal lobe, precentral gyrus, paracentral lobule, parietal lobe, cingulate gyrus, ventral diencephalon, choroid plexus, corpus callosum, pituitary gland, and additional subcortical and cortical regions. Each structure can be further subdivided, and corresponding absolute volumes, relative volumes, and derived indices are generated. In total, the system outputs 283 quantitative imaging indicators (Fig. 2).

Fig. 2.

Visualization of brain structure localization using the AI-based software

The segmentation performance of uAIDiscoverCSVD has been validated in previous studies, demonstrating good agreement with expert manual segmentation. Reported Dice similarity coefficients for major brain structures typically exceed 0.85, with low volumetric error for ventricular and cortical regions.

All automatically generated segmentation results in the present study underwent visual quality control by two experienced neuroradiologists who were blinded to clinical information. Images with evident segmentation errors or artifacts were excluded from subsequent analyses.

Absolute volume refers to the directly measured volume of a given brain structure derived from MRI-based segmentation and is expressed in cubic millimeters (mm³). Intracranial volume (ICV) was automatically computed by the software as the sum of gray matter, white matter, and cerebrospinal fluid volumes. Relative volume was calculated as the ratio of the absolute volume of a specific structure to the total ICV and expressed as a percentage (%), thereby accounting for inter-individual differences in head size.

Following preliminary screening for group differences and based on prior literature identifying brain regions potentially relevant to CSVD-related gait impairment, 21 relative brain structure volume indicators were selected for subsequent principal component and regression analyses.

For paired brain structures (e.g., hippocampus, thalamus, frontal lobe, temporal lobe), the reported volumes represent the combined volumes of the left and right hemispheres automatically calculated by the segmentation system. The present study primarily focused on global structural volume patterns rather than hemispheric asymmetry. Therefore, asymmetry indices between the left and right sides were not separately analyzed in the current study.

Data analysis

PCA combined with logistic regression was performed on the relative volume data of brain structures for each group of subjects. PCA was initially applied to the relative volume data, extracting principal components with a cumulative variance contribution rate exceeding 70%. The gait disorder outcome was used as the response variable, and principal component regression analysis was conducted to identify key brain structure volume indicators that influence gait disorders. A p-value of < 0.05 was considered statistically significant.

To ensure evaluation discipline and reproducibility, the analysis protocol was designed to minimize information leakage and to maintain a transparent separation between feature construction and outcome modeling. Dimensionality reduction using PCA was performed exclusively on normalized imaging-derived features without incorporating outcome labels, and subsequent regression analyses were conducted using fixed component scores. All preprocessing steps, including intracranial volume normalization, variable selection, standardization, and rotation strategy, were explicitly specified to facilitate reproducibility, in accordance with recent recommendations on transparent and leakage-safe analytical pipelines in biomedical imaging studies.

To evaluate whether principal components derived from brain structural volumes provide incremental information beyond conventional imaging markers, additional logistic regression models were constructed. In these models, principal component scores were entered together with established clinical imaging markers, including the Evans index and periventricular Fazekas score, which showed significant group differences in univariate analyses.

To enhance interpretability of the principal components, an orthogonal varimax rotation with Kaiser normalization was applied after component extraction. Rotation was used solely to achieve a clearer loading structure and did not alter the total variance explained by the retained components, although variance redistribution across individual components occurred. Importantly, rotated principal component scores generated by SPSS were used consistently in subsequent logistic regression analyses. Thus, component loadings, variance summaries, and regression models were all based on the same rotated component solution.

Principal component analysis was performed using the PCA extraction method in SPSS. As part of the standard PCA output, communalities (extracted) were examined to assess the proportion of variance in each original variable accounted for by the retained components. These values reflect how well the selected principal components represent each brain structure and should not be interpreted as common factor variances derived from exploratory factor analysis.

Prior to principal component analysis, all relative brain volume variables were mean-centered and scaled to unit variance (z-score standardization). Although all indicators were expressed as relative volumes (%), their variances and distributions differed across brain structures. Standardization was therefore applied to ensure that each variable contributed equally to the PCA and to prevent components from being disproportionately driven by structures with larger variances. PCA was subsequently performed on the correlation matrix rather than the covariance matrix.

Given the significant age difference between the GD and NGD groups and the known influence of age on ventricular size and cerebrospinal fluid spaces, age was included as a covariate in an adjusted logistic regression model. Principal components derived from relative brain volumes were entered simultaneously with age to assess whether the associations between structural volume patterns and gait disturbance remained independent of age. In additional exploratory models, sex, years of education, and vascular risk factors (hypertension, diabetes, smoking, and coronary heart disease) were also entered as covariates to evaluate the robustness of the principal component effects. Due to the moderate sample size, these multivariable-adjusted models were considered supportive analyses rather than primary models.

Prior to principal component analysis (PCA), all relative brain volume variables were standardized using z-score normalization to ensure comparability across structures with different scales. The suitability of the dataset for PCA was evaluated by examining the communality (extracted)s and the overall variance explained by extracted components. Only components with eigenvalues greater than 1 and a cumulative variance contribution exceeding 70% were retained for subsequent analyses, in accordance with established methodological recommendations. To reduce multicollinearity and overfitting risks in regression modeling, principal components rather than individual brain volume measures were entered into the logistic regression model. This dimensionality reduction strategy was chosen to ensure statistical stability given the moderate sample size and the high correlation among regional brain volumes. All regression models were checked for convergence and stability, and odds ratios with 95% confidence intervals were reported to facilitate clinical interpretation.

Because the primary aim of this study was to explore structural brain volume patterns associated with gait disorders in patients with CSVD rather than to develop a predictive classification model, the dataset was not divided into separate training and validation cohorts. Given the relatively limited sample size (N = 90), splitting the data could substantially reduce statistical power and stability of the PCA results. Therefore, principal component analysis was performed on the full dataset to identify dominant structural patterns, and logistic regression was subsequently used to examine their association with gait disorder status. Future studies with larger cohorts may consider independent validation datasets to further evaluate the robustness and generalizability of the findings.

Results

Demographic data

A total of 90 patients were included in the study, comprising 48 men and 42 women. Among them, 40 patients were classified into the non-gait disorder (NGD) group (21 men and 19 women), and 50 patients into the gait disorder (GD) group (27 men and 23 women). As summarized in Table 1, no significant differences were observed between the GD and NGD groups with respect to sex distribution, years of education, or vascular risk factors, including smoking status, hypertension, diabetes mellitus, and coronary heart disease (all P > 0.05). In contrast, age differed significantly between the two groups (P = 0.012), with patients in the GD group being older than those in the NGD group.

Table 1.

Comparison of Demographic Data and Risk Factors

Demographic Data	Group NGD(N = 40)	Group GD (N = 50)	P
Age (years)	68.5(64,71)	70.5(66.75,75)	0.012
Gender (Male/Female)	19/21	27/23	0.887
Education Years (years)	9 (9,9)	9 (9,9)	0.842
Risk Factors
Smoking, N (%)	11 (27.5)	15 (30.0)	0.795
Hypertension, N (%)	27 (67.5)	30 (60.0)	0.463
Diabetes, N (%)	13 (32.5)	13 (26.0)	0.499
Coronary Heart Disease, N (%)	8 (20.0)	9 (18.0)	0.810

Note: NGD = non-gait disorder group; GD = gait disorder group; N = number of participants; % = percentage. Data are presented as median (interquartile range) or number (percentage). Group comparisons were performed using the Mann–Whitney U test for continuous variables and the χ² test for categorical variables. A two-tailed P value < 0.05 was considered statistically significant

MRI imaging indicators

As shown in Table 2, there were no significant differences between the GD and NGD groups in terms of the total burden score (P = 0.487) and deep white matter Fazekas score (P > 0.05). However, significant differences were observed in the bilateral MTA score, paraventricular Fazekas score, and Evans index (P < 0.05), indicating that the GD group exhibits more severe brain atrophy, paraventricular white matter lesions, and ventricular system enlargement.

Table 2.

Comparison of imaging scores and evans index between the NGD and GD groups (N = 90)

Imaging Score and Evans Index	Group NGD(N = 40)	Group GD (N = 50)	P
Total burden score	2(2,3)	2(2,3)	0.487
Left MTA score	0(0,1)	1(0,1)	0.004
Right MTA score	0(0,0.75)	0(0,1)	0.037
Periventricular Fazekas score	2(2,3)	3(2,3)	0.013
Deep white matter Fazekas score	3(2,3)	3(2,3)	0.068
Evans index	0.27(0.25,0.29)	0.28(0.27,0.30)	0.001

Note: Values are presented as median (interquartile range). Comparisons between groups were performed using the Mann–Whitney U test for continuous or ordinal variables. A two-tailed P value < 0.05 was considered statistically significant

Brain structural volume

The volume distribution of brain structures reflects the structural differences in brain tissue between patients with small vessel disease and gait disorders. As shown in Tables 3 and 4, the volumes of various brain structures differed to varying extents between the GD and NGD groups. Significant differences in both absolute and relative volumes were observed in several brain structures, including total cerebrospinal fluid, peripheral cerebrospinal fluid, third ventricle, lateral ventricle, thalamus, precentral gyrus, cingulate gyrus, ventral diencephalon, and choroid plexus. However, in some structures (e.g., hippocampus, globus pallidus, temporal lobe, frontal lobe, corpus callosum), although absolute volume differences were not significant, relative volume changes were more pronounced.

Table 3.

Comparison of absolute brain volumes between NGD and GD groups (N = 90)

Absolute brain volume (cm³)	NGD (n = 40)	GD (n = 50)	P value
Total brain volume	1427.44 ± 134.51	1450.19 ± 117.47	0.394
Total cerebrospinal fluid (CSF)	357.70 (320.42, 390.06)	403.65 (366.53, 433.34)	< 0.001
Peripheral CSF	322.86 (289.53, 352.49)	342.43 (321.87, 377.91)	0.002
Third ventricle	1.69 (1.39, 2.34)	2.34 (1.98, 2.67)	< 0.001
Fourth ventricle	1.93 (1.58, 2.39)	2.08 (1.73, 2.31)	0.626
Lateral ventricles	30.18 (21.15, 39.34)	42.81 (32.62, 58.59)	< 0.001
Hippocampus	6.42 (5.99, 7.01)	6.14 (5.80, 6.61)	0.089
Globus pallidus	3.41 (3.03, 3.76)	3.29 (3.11, 3.53)	0.314
Thalamus	12.19 (11.37, 13.65)	11.24 (10.48, 12.96)	0.03
Temporal lobe	12.59 (11.65, 14.02)	12.74 (11.80, 14.03)	0.865
Frontal lobe	72.96 (68.57, 81.04)	72.89 (65.40, 78.40)	0.426
Precentral gyrus	102.00 ± 12.15	99.17 ± 11.19	0.254
Cingulate gyrus	25.00 ± 2.35	23.62 ± 2.44	0.008
Ventral midbrain	6.91 (6.67, 7.36)	6.71 (6.32, 7.39)	0.211
Choroid plexus	23.81 ± 2.83	23.45 ± 2.63	0.538
Corpus callosum	17.77 ± 2.43	16.59 ± 1.91	0.012

Footnote: Values are presented as mean ± SD or median (interquartile range), as appropriate. Absolute brain volumes are expressed in cubic centimeters (cm³)

Table 4.

Comparison of relative brain volumes between NGD and GD groups (N = 90)

Relative brain volume (%)	NGD (n = 40)	GD (n = 50)	P value
Total cerebrospinal fluid (CSF)	25.07 ± 3.13	28.17 ± 2.99	< 0.001
Peripheral CSF	22.44 ± 2.54	24.30 ± 2.23	< 0.001
Third ventricle	0.12 (0.10, 0.17)	0.16 (0.14, 0.18)	< 0.001
Fourth ventricle	0.14 ± 0.04	0.14 ± 0.03	0.876
Lateral ventricles	2.10 (1.45, 2.61)	2.94 (2.36, 4.17)	< 0.001
Hippocampus	0.46 (0.43, 0.49)	0.43 (0.40, 0.46)	0.008
Globus pallidus	0.24 (0.22, 0.26)	0.23 (0.22, 0.24)	0.028
Thalamus	0.88 (0.80, 0.97)	0.79 (0.71, 0.88)	0.003
Insula	0.89 (0.84, 0.96)	0.87 (0.83, 0.91)	0.137
Temporal lobe	5.16 (5.02, 5.43)	5.03 (4.78, 5.22)	0.014
Frontal lobe	7.15 ± 0.50	6.85 ± 0.53	0.006
Precentral gyrus	1.76 ± 0.12	1.63 ± 0.15	< 0.001
Paracentral lobule	0.50 ± 0.05	0.48 ± 0.05	0.065
Superior parietal lobule	1.67 ± 0.13	1.62 ± 0.16	0.128
Cingulate gyrus	1.24 ± 0.12	1.14 ± 0.11	< 0.001
Ventral midbrain	0.31 (0.29, 0.33)	0.28 (0.26, 0.30)	< 0.001
Choroid plexus	0.15 (0.11, 0.18)	0.19 (0.15, 0.21)	< 0.001
Corpus callosum	0.26 ± 0.04	0.24 ± 0.04	0.03
Pituitary+	0.04 ± 0.01	0.03 ± 0.01	0.57

Footnotes: Relative brain volumes are expressed as percentages (%) of intracranial volume. +The pituitary is shown for completeness but was not included among the 21 relative-volume indicators selected for principal component analysis

PCA-Logistic regression analysis of relative volume of brain structures

As shown in Table 5, the communalities (extracted) for all brain structures exceeded 0.5, indicating that a substantial proportion of variance in each variable was accounted for by the retained principal components. Brain structures with higher communalities, such as cerebrospinal fluid (0.833), hippocampus (0.841), and ventral diencephalon (0.931), were well represented in the PCA solution, suggesting that their variability was largely captured by the extracted components. These values reflect the adequacy of the PCA model in representing the selected brain structures, rather than implying differential importance of individual regions in subsequent analyses.

Table 5.

Communality (extracted) for each brain structure (relative volume, N = 90)

Brain Region	Communality (extracted)	Brain Region	Communality (extracted)
Cerebrospinal Fluid	0.830	Temporal Lobe	0.822
Third Ventricle	0.633	Precentral Gyrus	0.658
Fourth Ventricle	0.617	Paracentral Lobule	0.845
Lateral Ventricle	0.657	Superior Parietal Lobule	0.626
Hippocampus	0.841	Cingulate Gyrus	0.700
Globus Pallidus	0.649	Ventral Midbrain	0.931
Thalamus	0.828	Choroid Plexus	0.536
Insula	0.734	Corpus Callosum	0.707
Frontal Lobe	0.802	Pituitary	0.780

It should be emphasized that the sign of principal components is mathematically arbitrary in PCA and does not carry intrinsic biological directionality. Therefore, the interpretation of the regression coefficients for PC2 and PC3 was not based on component sign alone, but rather on the loading patterns of the original variables and their group-wise differences between GD and NGD participants. To provide a more interpretable link between principal components and clinical phenotype, we examined the group-wise differences of the top-loading original variables within PC2 and PC3. Variables with the largest absolute loadings on PC2 included cerebrospinal fluid, frontal lobe, temporal lobe, and cingulate gyrus, whereas PC3 was primarily characterized by high loadings for the third ventricle, lateral ventricles, and corpus callosum. These variables represent the main structural contributors to the respective principal components. Because the sign of PCA loadings is mathematically arbitrary, interpretation focused on the magnitude of the loadings rather than their direction. Specifically, cerebrospinal fluid volume and ventricular volumes were significantly higher in the GD group, whereas cortical and commissural structures exhibited lower relative volumes. Standardized mean differences between GD and NGD groups for these variables were in directions consistent with worse gait performance in the GD group. These findings indicate that the association between PC2/PC3 and gait disturbance reflects coherent patterns of ventricular expansion and cortical-white matter atrophy, rather than an artifact of component sign orientation.

When the Evans index and periventricular Fazekas score were included in the regression model together with the principal components, Components 2 and 3 remained significantly associated with gait disturbance. The effect estimates for PC2 and PC3 were only modestly attenuated, indicating that the structural volume patterns captured by these components provide information beyond that conveyed by standard imaging markers alone. In contrast, the total CSVD burden score did not show a significant association with gait disturbance in either unadjusted or adjusted models and did not materially affect the associations observed for PC2 and PC3.

To assess the robustness of the results to the choice of TUG cutoff, sensitivity analyses were performed using alternative thresholds of 10.5 s and 12.5 s to define gait disturbance. Across these alternative definitions, the overall pattern of results remained consistent. In particular, principal components reflecting cerebrospinal fluid expansion and ventricular enlargement (Components 2 and 3) remained significantly associated with gait impairment, with comparable effect directions and magnitudes. Although minor variations in statistical significance were observed for components with weaker effects, the main conclusions were not materially altered by the choice of cutoff.

It should be noted that the identified principal components represent patterns of covarying brain structure volumes rather than isolated regional effects. Therefore, the regression results should be interpreted as associations between structural volume patterns and gait disturbance, rather than as evidence of direct causal relationships. Given the cross-sectional nature of the study, the observed associations reflect concurrent structural-functional relationships and do not imply temporal precedence. These findings nonetheless provide clinically meaningful insight into neuroanatomical configurations that are most strongly linked to gait impairment in CSVD.

As shown in Table 6, Component 1 exhibits the highest variance contribution, accounting for 41.14% of the relative volume variation in brain structures, making it the most significant principal component. Component 2 follows with a variance contribution rate of 12.61%, explaining the second-largest portion of the variation. Although Components 3, 4, and 5 have lower variance contributions, they still provide meaningful explanatory power. The variance contribution rates reported after rotation (rotational load sum of squares) reflect the redistributed variance across components following varimax rotation and were used to describe the relative importance of each rotated component. Interpretation of component meaning was therefore based on rotated loadings rather than on the unrotated eigenvalue structure.

Table 6.

Variance contribution rate of each component (relative volume, N = 90)

Component	Initial Eigenvalue			Rotational Load Sum of Squares
	Eigenvalue	Variance Percentage	Cumulative Percentage	Eigenvalue	Variance Percentage	Cumulative Percentage
1	7.406	41.144	41.144	3.735	20.748	20.748
2	2.270	12.611	53.755	3.675	20.415	41.163
3	1.248	6.932	60.687	2.682	14.901	56.064
4	1.189	6.603	67.290	1.913	10.629	66.693
5	1.083	6.014	73.304	1.190	6.611	73.304
6	0.821	4.561	77.865
7	0.733	4.072	81.937
8	0.626	3.479	85.416
9	0.544	3.021	88.437
10	0.495	2.748	91.185
11	0.399	2.219	93.404
12	0.269	1.494	94.897
13	0.233	1.296	96.193
14	0.196	1.090	97.283
15	0.184	1.021	98.305
16	0.141	0.782	99.087
17	0.102	0.569	99.656
18	0.062	0.344	100.000

Principal components were retained according to the criterion that the cumulative explained variance exceeded 70%. Based on this rule, the first five principal components (PC1-PC5) were extracted, which together explained the majority of the total variance of the included brain structure variables. Components beyond PC5 contributed only marginal additional variance and were therefore not retained for further analysis. In Table 6, only the extracted components (PC1–PC5) have corresponding eigenvalues and variance statistics reported. Components 6–18 were not retained in the final PCA solution, and therefore their eigenvalues and variance percentages are not presented in the extraction results.

As shown in the scree plot (Fig. 3), the characteristic values of Component 1 and Component 2 are significantly higher than those of the other components, indicating that these two components offer substantial discriminatory power in principal component analysis. This suggests they may capture the primary patterns of brain structure volume changes. In this study, components with a cumulative variance contribution rate greater than 70% were selected for further analysis, specifically the first to fifth components.

Fig. 3.

Scree plot of variance contribution rate (relative volume, N = 90)

The rotated component load matrix helps to simplify the structure and enhance interpretability. As shown in Table 7, the hippocampus (0.915), globus pallidus (0.769), thalamus (0.765), insular lobe (0.765), and ventral diencephalon (0.800) exhibit higher loadings on Component (1) The cerebrospinal fluid (CSF) (-0.606), temporal lobe (0.804), frontal lobe (0.845), and cingulate gyrus (0.692) have higher loadings on Component (2) The third ventricle (0.629), lateral ventricle (0.618), and corpus callosum (0.813) load most heavily on Component (3) The paracentral lobule (0.895) and superior lobule (0.683) show stronger loadings on Component 4, while the pituitary gland (0.858) exhibits a high loading on Component 5. The fourth ventricle shows a slightly higher loading on both Component 2 (-0.547) and Component 5 (0.526).

Table 7.

Rotated component loading matrix* (relative volume, N = 90)

	Component 1	Component 2	Component 3	Component 4	Component 5
Cerebrospinal Fluid	-0.347	-0.606	-0.503	-0.300	-0.014
Third Ventricle	-0.192	-0.276	-0.629	-0.349	0.058
Fourth Ventricle	-0.015	-0.547	-0.191	-0.057	0.526
Lateral Ventricle	-0.159	-0.420	-0.618	-0.250	-0.103
Hippocampus	0.915	0.031	0.021	0.052	-0.004
Globus Pallidus	0.769	0.091	0.221	-0.004	0.015
Thalamus	0.765	0.176	0.435	0.130	-0.078
Insula	0.765	0.359	-0.022	0.135	-0.038
Temporal Lobe	0.354	0.804	0.028	-0.049	0.162
Frontal Lobe	0.076	0.845	0.051	0.307	0.070
Precentral Gyrus	0.180	0.555	0.350	0.427	0.114
Paracentral Lobule	0.168	0.044	0.066	0.895	0.102
Superior Parietal Lobule	-0.051	0.305	0.154	0.683	-0.201
Cingulate Gyrus	0.085	0.692	0.409	0.212	0.038
Ventral Midbrain	0.800	0.192	0.499	0.065	0.013
Choroid Plexus	-0.262	-0.517	-0.374	0.047	0.240
Corpus Callosum	0.206	0.044	0.813	-0.032	-0.024
Pituitary	-0.032	0.195	0.067	-0.006	0.858

*: The rotation method used was Kaiser Normalization with maximum variance

In principal component analysis, each principal component represents a linear combination of all standardized variables included in the model. Therefore, PC1-PC5 are mathematically defined as weighted sums of the relative brain volume variables rather than being composed exclusively of a few structures. The score of a principal component can be expressed in the following general form:

where represent the standardized relative volumes of the included brain structures and w₁–wₙ denote the corresponding component loadings obtained from PCA. In practice, interpretation of principal components typically focuses on variables with the largest absolute loadings because these variables contribute most strongly to the component structure. In the present study, PC2 was mainly characterized by high loadings of cerebrospinal fluid, frontal lobe, temporal lobe, and cingulate gyrus, whereas PC3 showed the strongest loadings for the third ventricle, lateral ventricles, and corpus callosum. PC1 primarily reflected covarying volumetric changes in subcortical and limbic structures, including the hippocampus, globus pallidus, thalamus, and ventral diencephalon.

As shown in Table 8, a logistic regression analysis was performed using CSVD gait disturbance as the outcome variable, with Components 1 to 5 as predictors, to assess the predictive ability of these principal components for gait disturbances in CSVD. Component 1 primarily includes basal ganglia and medial temporal lobe structures, such as the hippocampus, globus pallidus, and thalamus, which are closely associated with memory and motor control. Component 5 mainly consists of the pituitary gland, which may be related to endocrine function. The regression coefficients for Component 1 and Component 5 were not statistically significant (P > 0.1), indicating that these components have a weak predictive ability for gait disorders in CSVD. Component 4, which includes the paracentral lobule and superior lobule, areas associated with sensorimotor function, had a regression coefficient of -0.456 (P = 0.075), approaching but not reaching statistical significance (P < 0.1), suggesting a modest negative predictive effect. The fourth ventricle, which had a slightly higher loading on Component 5 (0.526), did not significantly affect Component 5’s predictive power, suggesting that the fourth ventricle volume has a weak predictive relationship with gait disorders in CSVD. Component 2 (regression coefficient = -0.747, P = 0.010) involves cortical areas such as cerebrospinal fluid, temporal lobe, and frontal lobe, reflecting brain atrophy and changes in the ventricular system. Component 3 (regression coefficient = -1.121, P < 0.001) includes the third ventricle, lateral ventricle, and corpus callosum, which may be related to ventricular dilation and the integrity of white matter fibers. Both Components 2 and 3 demonstrate significant negative predictive capabilities. Based on the positive and negative component loadings, the increase in cerebrospinal fluid, ventricular system expansion, and atrophy of cortical areas such as the temporal lobe, frontal lobe, precentral gyrus, and corpus callosum are significant predictors of gait disorders in CSVD.

Table 8.

PCA-logistic regression analysis results (relative volume, N = 90)

Component*	Regression Coefficient	Standard Error	Wald	P	Exp(B)	Exp(B) (95% Confidence Interval)
						Lower Limit	Upper Limit
Component 1	-0.049	0.249	0.039	0.844	0.952	0.584	1.552
Component 2	-0.747	0.291	6.589	0.010	0.474	0.268	0.838
Component 3	-1.121	0.310	13.066	<0.001	0.326	0.178	0.599
Component 4	-0.456	0.256	3.161	0.075	0.634	0.383	1.048
Component 5	-0.357	0.260	1.875	0.171	0.700	0.420	1.166
Constant	0.284	0.250	1.299	0.254	1.329

Note: CI = confidence interval; Exp(B) = odds ratio (OR). The results were obtained from logistic regression analysis based on principal component scores

*: Component 1 mainly includes the hippocampus, globus pallidus, and thalamus. Component 2 mainly involves cerebrospinal fluid, temporal lobe, frontal lobe, and cingulate gyrus. Component 3 mainly consists of the third ventricle, lateral ventricle, and corpus callosum. Component 4 mainly includes the paracentral lobule and superior parietal lobule. Component 5 primarily involves the pituitary

Note: Exp(B) represents the odds ratio (OR). CI indicates confidence interval. Logistic regression analysis was performed using the principal component scores as predictors. A two-tailed P value < 0.05 was considered statistically significant

After adjusting for age, the associations between gait disturbance and principal components reflecting cerebrospinal fluid and ventricular enlargement remained statistically significant. Specifically, Component 2 (adjusted OR < 1, P < 0.05) and Component 3 (adjusted OR < 1, P < 0.01) continued to show independent associations with gait disturbance. The direction and magnitude of the effect estimates were comparable to those observed in the unadjusted model, indicating that the observed relationships were not solely driven by age-related structural changes. Further adjustment for sex, years of education, and vascular risk factors did not materially alter the associations for Components 2 and 3, although confidence intervals widened slightly, consistent with the limited sample size.

Discussion

Influence of demographic factors on gait disorders in CSVD

Among the patients included in this study, the GD group was older than the NGD group, suggesting that the risk of CSVD increases with age and that age may be an independent risk factor for gait disorders. This finding aligns with existing literature, which links age-related vascular changes, such as vascular stiffening and blood-brain barrier disruption, to white matter lesions [10], however, this age difference may also be influenced by selection bias in the study population, with older patients being more likely to be included, rather than the difference being solely attributed to the disease itself. There were no significant differences between the two groups in terms of gender or years of education, though these factors should not be entirely excluded. For instance, years of education may impact cognitive function through a cognitive reserve mechanism, which has been linked to gait disorders [11]. Additionally, no significant differences were found between the two groups in risk factors such as smoking, hypertension, diabetes, or coronary heart disease, suggesting that the pathogenesis of CSVD likely involves a complex interplay of multiple factors, with any single risk factor insufficient to explain the clinical manifestations of the disease.

Predictive effect of CSVD imaging score on gait disorders

This study demonstrates that the CSVD total load score does not effectively predict gait disorders, which contrasts with findings from some studies that suggest the total load score is associated with motor function [12]. Possible explanations for this discrepancy include: (1) The total load score is typically based on a simple aggregation of the number or volume of lacunar infarcts, white matter hyperintensities (WMH), and other lesions. However, differences in the anatomical location and functional connectivity of these lesions may have heterogeneous effects on gait [13]. For instance, in this study, an increase in the volume of paraventricular WMH was significantly associated with gait disturbance, while no significant difference was observed in deep white matter WMH, highlighting the importance of lesion location. (2) The volume characteristics of brain structures (such as the ventricular system and corpus callosum), as extracted by the intelligent image analysis system through principal component analysis (PCA), may more sensitively reflect the pathological progression of CSVD. Recent research has emphasized that multi-modal, image-based models of local network injury are superior to traditional scoring systems in predicting motor dysfunction [14], which supports the findings of this study. Moving forward, developing a more accurate CSVD neurological injury evaluation system, based on advancements in imaging research, should be prioritized.

The Evans index, bilateral MTA score, and paraventricular Fazekas score in the GD group were significantly higher than those in the NGD group, while no significant difference was observed in the deep white matter Fazekas score. These findings suggest that brain atrophy and paraventricular white matter lesions are closely associated with gait disorders. Brain atrophy may impair the motor cortex and related neural pathways, leading to reduced motor function, while paraventricular white matter lesions could disrupt nerve conduction, further impacting gait control [15]. For instance, damage to the fibers connecting the motor cortex and basal ganglia can affect gait by disrupting the cortex-basal ganglia-cerebellar circuit [16]. The persistence of PC2 and PC3 associations after adjustment for the Evans index and periventricular Fazekas score suggests that multiregional structural volume patterns capture aspects of gait-related brain vulnerability that are not fully reflected by conventional imaging markers. While ventricular enlargement and periventricular white matter lesions are clinically informative, the PCA-derived components integrate distributed cortical, subcortical, and ventricular changes into coherent patterns, thereby offering incremental value for understanding gait impairment in CSVD.

Differences in brain structure and volume in CSVD gait disorders

Absolute volume directly reflects the actual size changes of brain structures, while relative volume more accurately captures the proportional relationship between brain structures. In this study, there were notable differences in both the absolute and relative volume distributions of certain brain structures between the two groups. In the GD group, changes in relative volume were more pronounced, suggesting that relative volume is a more sensitive indicator. Several factors may contribute to this observation: (1) Patients with CSVD often exhibit brain atrophy and ventricular enlargement [17], leading to a significant increase in the absolute volume of the ventricular system and cerebrospinal fluid, while the overall brain volume remains relatively stable. (2) The relative stability of certain gray matter structures. Some gray matter structures, such as the thalamus, may be less affected in CSVD [10], which may explain why their absolute volume changes are not as significant. (3) White matter structures, such as the corpus callosum, are often significantly impacted in CSVD [18], and thus the changes in their relative volume are more pronounced.

Neuroanatomical factors for gait disorders in CSVD

The results of this study indicate that gait disorders in patients with CSVD are closely associated with pathological changes such as brain atrophy, paraventricular white matter lesions, and ventricular enlargement, which aligns with existing research [19]. This further supports the link between abnormal cerebrospinal fluid dynamics and motor impairment [20]. Previous studies have demonstrated that brain atrophy and paraventricular white matter lesions are strongly correlated with gait disorders. Brain atrophy may impact the motor cortex and related neural pathways, leading to reduced motor function, while paraventricular white matter lesions can disrupt nerve conduction, thereby affecting gait control [15]. Increased white matter hyperintensities and paraventricular white matter lesions may result in damage to white matter fibers, including those connecting the motor cortex and basal ganglia, ultimately impairing gait control by disrupting the cortex-basal ganglia-cerebellar circuit [16]. Changes in the ventricular system, such as ventricular dilation and periventricular white matter lesions, are hallmark imaging features of CSVD. Clinically, periventricular white matter lesions often lead to an unstable gait and slow walking [4]. Ventricular dilation and abnormal cerebrospinal fluid circulation may represent critical pathophysiological mechanisms underlying gait disorders in CSVD, potentially causing mechanical compression of brain tissue and metabolic disruptions that affect gait function. This finding offers a new perspective for the clinical diagnosis and treatment of CSVD, with future research needed to explore the pathophysiological mechanisms of the ventricular system and potential intervention strategies. Additionally, studies have shown that the clinical triad of gait disorders, cognitive impairment, and bladder dysfunction is characteristic of idiopathic normal pressure hydrocephalus [21, 22].Therefore, patients with CSVD-related gait disorders should be assessed for hydrocephalus, and the specific neuropathological mechanisms warrant further investigation. Accordingly, statements regarding “positive” or “negative” effects should be understood as referring to the direction of group-level structural differences rather than to the sign of principal component scores. The consistent enlargement of cerebrospinal fluid spaces and ventricles, together with relative atrophy of frontal, temporal, and commissural structures in the GD group, provides the anatomical basis for the observed associations with gait impairment. Importantly, although ventricular enlargement and increased cerebrospinal fluid spaces are strongly age-dependent processes, the persistence of these associations after age adjustment suggests that the identified structural patterns capture disease-related vulnerability beyond normal aging. These findings indicate that age alone does not fully explain the relationship between CSVD-related structural changes and gait impairment.

Shortcomings and prospects

The limitations of this study are primarily reflected in the following aspects: First, the sample size is relatively small, which may result in sample imbalance and insufficient statistical power for certain findings. Second, the study design is cross-sectional, lacking long-term follow-up data. As a result, the causal relationship between imaging changes and gait disorders remains unclear, and the study cannot assess the natural progression of CSVD or the dynamic changes and long-term prognosis of gait disorders. Third, the study is single-center, which introduces potential selection bias and limits the generalizability and representativeness of the results. Future studies with larger sample sizes, long-term follow-up, and multi-center designs are needed to validate and expand upon these findings.

Future research should focus on expanding the sample size, particularly when conducting principal component analysis (PCA). A larger cohort will enhance the predictive power of the principal components regarding CSVD incidence and facilitate the development of early screening tools or risk assessment models for CSVD. Longitudinal studies, integrated with multimodal imaging technologies, are also needed to validate causal relationships and explore targeted intervention strategies. These studies could investigate whether interventions focused on the cerebral cortex and ventricular system can mitigate the risk of CSVD progression. Moreover, future research should consider expanding imaging modalities. Although traditional MRI is effective in detecting many characteristics of CSVD, its sensitivity to subtle pathological changes, such as microvascular lesions and blood-brain barrier integrity, is limited [8, 23]. Emerging MRI techniques, such as dynamic contrast-enhanced MRI (DCE-MRI), dynamic susceptibility contrast MRI (DSC-MRI), and high-field MRI (e.g., 7T MRI), will provide more comprehensive insights into brain tissue metabolism, blood flow perfusion, and microvascular structures. Additionally, diffusion tensor imaging (DTI) can assess white matter integrity and reveal ultrastructural damage. In the future, DTI could serve as a valuable surrogate marker for white matter fiber damage in clinical research, aiding in the early detection and monitoring of CSVD [24]. Although the sample size was sufficient for exploratory PCA and multivariable regression using a reduced set of principal components, future studies with larger, multicenter cohorts will be necessary to further validate the stability and generalizability of the identified component structure.

The findings of the present study are generally consistent with previous research investigating the relationship between structural brain alterations and gait impairment in patients with cerebral small vessel disease. Prior studies have also reported that ventricular enlargement, increased cerebrospinal fluid spaces, and cortical atrophy are frequently associated with gait disturbances and reduced mobility in individuals with small vessel disease. These structural changes may reflect diffuse brain tissue loss and disruption of motor-related neural networks, which can ultimately affect gait performance and balance control. Therefore, the results of this study further support the notion that global brain structural changes play an important role in the development of gait disorders in patients with cerebral small vessel disease.

Conclusion

In this cross-sectional study, gait disturbance in patients with cerebral small vessel disease was found to be significantly associated with enlargement of cerebrospinal fluid spaces and the ventricular system, as well as with relative atrophy of cortical and commissural structures, including the temporal lobe, frontal lobe, and corpus callosum. These structural alterations appear to co-occur as part of a broader pattern of brain tissue loss and network disruption, rather than representing isolated causal drivers of gait impairment. Notably, relative volume changes in these regions were more pronounced in patients with gait disturbance, whereas the total CSVD burden score showed limited ability to reflect gait-related functional impairment. This suggests that multiregional volumetric patterns may capture aspects of disease-related vulnerability that are not fully represented by conventional aggregate scores. Ventricular enlargement in CSVD should therefore be interpreted as an imaging marker of diffuse tissue loss or co-existing pathological processes, rather than as evidence of a direct causal mechanism. Although patients with hydrocephalus were excluded from the present study, recognition of gait disturbance, cognitive impairment, and ventricular enlargement remains clinically relevant for differential diagnosis, including consideration of idiopathic normal pressure hydrocephalus in appropriate contexts. Future studies with larger sample sizes, longitudinal designs, and multimodal imaging approaches are needed to clarify temporal relationships and underlying mechanisms linking structural brain changes to gait dysfunction in CSVD.

Abbreviations

CSVD

Cerebral small vessel disease

GD

Gait disorders

NGD

Non-gait disorder

PCA

Principal component analysis

TUG

Timed up and go test

MRI

Magnetic resonance imaging

CMB

Cerebral microbleeds

mRS

Modified rankin scale

T1W1

T1-weighted imaging

T2W1

T2-weighted imaging

FLAIR

Fluid-attenuated inversion recovery

DWI

Diffusion-weighted imaging

SWI

Susceptibility-weighted imaging

CI

Confidence interval

CSF

Cerebrospinal fluid

WMH

White matter hyperintensities

DCE-MRI

Dynamic contrast-enhanced MRI

DSC-MRI

Dynamic susceptibility contrast MRI

Author contributions

Hong-Bo Wang: Writing Hua-Wei Chen: Data analysis Xiao-Ying Zhao: Resources Jia-Jing Tian: Visualization Bing-Bing Gao: Data curation Yan-Wei Miao: Project administration Bing-Wei Zhang: Project administration and funding acquisition All authors read and approved the final manuscript.

Funding

This study was supported by: (1) Liaoning Provincial Science and Technology Program Joint Plan - General Program of Natural Science Foundation (No. 2024-MSLH-082); (2) Xingliao Talents Program - Distinguished Medical Experts (XLYC2412084).

Data availability

Full data may be shared with any qualified researcher upon reasonable request, subject to consent from both the patients and the Hospital Ethics Committee. (Contact: Hua-Wei Chen ✉chenhuawei2020@163.com).

Declarations

Ethics approval and consent to participate

The experimental protocol was established, according to the ethical guidelines of the Helsinki Declaration and was approved by the Bioethics Committee of the First Affiliated Hospital of Dalian Medical University (reference number: YJ-KS-KY-2024-318). Written informed consent was obtained from individual or guardian participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.