Introduction
Ischemic cerebrovascular disease arises from heterogeneous vascular processes, including large-artery atherosclerosis, acute small perforator infarction, and chronic small-vessel–mediated brain injury[
1–
3]. Although these conditions share conventional vascular risk factors, they differ in vascular territory, lesion distribution, temporal profile, and downstream tissue injury. Magnetic resonance imaging (MRI) remains central to cerebrovascular phenotyping, particularly for detecting recent small subcortical infarcts (RSSI), white matter hyperintensities (WMH), lacunes, enlarged perivascular spaces, cerebral microbleeds, and brain atrophy[
1]. However, despite its indispensable role, MRI-based assessment is not always readily scalable because of cost, scanner availability, acquisition burden, contraindications, and, in selected vascular assessments, the need for contrast agents or invasive angiography[
4]. Accessible imaging markers that can complement MRI and capture mechanism-relevant vascular injury are therefore needed.
The retina offers a biologically plausible and clinically accessible window into cerebrovascular health[
5–
7]. As an extension of the central nervous system, the retina shares developmental, anatomical, and microvascular features with the brain and can be imaged rapidly and noninvasively. Optical coherence tomography (OCT) provides quantitative measurements of retinal neural layers, whereas optical coherence tomography angiography (OCTA) enables depth-resolved assessment of retinal and choroidal microvascular networks[
8]. Previous studies[
9–
12] have associated retinal structural and microvascular abnormalities with carotid artery stenosis, ischemic stroke subtypes, and cerebral small vessel disease (CSVD). In patients with carotid artery stenosis, OCTA studies[
9,
13] have reported altered retinal perfusion and vascular density, including changes after carotid revascularization. In ischemic stroke, retinal OCT/OCTA metrics have been shown to differ between large-artery atherosclerosis and small-artery disease[
11,
12]. In CSVD, retinal biomarkers have been associated with WMH burden and composite MRI markers of small-vessel injury[
5,
14].
Nevertheless, important uncertainties remain. Most prior studies[
9,
11,
14] have examined a single cerebrovascular condition or relied on pairwise comparisons, leaving unclear whether carotid stenosis-related stroke, RSSI, and chronic CSVD exhibit distinct retinal imaging signatures when evaluated within a unified analytical framework. In addition, many studies have focused on visual MRI ratings or isolated retinal parameters, whereas fewer have integrated OCT/OCTA measures with quantitative MRI-derived markers of brain tissue loss and white matter injury[
15]. More importantly, it remains uncertain whether retina–brain associations differ across vascular disease categories. Addressing this issue may help distinguish whether retinal imaging primarily reflects nonspecific vascular burden or provides mechanism-relevant information across different forms of cerebrovascular injury.
In this study, we compared retinal OCT/OCTA metrics and quantitative MRI-derived brain imaging markers among controls, patients with carotid stenosis-related stroke, patients with RSSI, and patients with chronic CSVD. We first evaluated group-specific retinal structural and microvascular profiles and then tested whether associations between OCTA-derived vascular metrics and MRI-derived gray matter volume, WMH burden, and regional brain volumes differed by disease category. We hypothesized that retinal imaging profiles and retina–brain associations would vary across cerebrovascular disease categories, reflecting distinct patterns of large-artery, acute small perforator, and chronic diffuse small-vessel injury.
Materials and methods
Study setting and participants
This registry-based cross-sectional study was conducted within the framework of the A Cohort Study of Neuroimaging, Ophthalmic Imaging, and Clinical Features in Cerebral Large Artery Stenosis (NOICAS registry), an ongoing observational multicenter study investigating associations between retinal imaging and neuroimaging in cerebrovascular disease (ChiCTR2300074640). The present analysis included patients with acute ischemic stroke admitted to the Departments of Neurology of three tertiary hospitals in Chengdu, China, and participants with chronic CSVD recruited from outpatient clinics and local communities between February 2021 and December 2024. Age- and sex-comparable adults aged > 50 years were selected as controls from the same source population whenever possible. Participants enrolled before registry registration were assessed using the same standardized imaging and clinical protocols. The study was approved by the Ethics Committee of West China Hospital, Sichuan University [approval number: 2020(922)]. Written informed consent was obtained from all participants, and the study was conducted in accordance with the Declaration of Helsinki.
All participants underwent brain MRI, OCT/OCTA imaging, neurological history taking, and physical examination. All MRI and OCT/OCTA examinations in the carotid artery stenosis-associated stroke (CAS) group were completed before carotid revascularization therapy. Demographic characteristics and vascular risk factors, including age, sex, hypertension, diabetes, dyslipidemia, smoking, and drinking, were collected using standardized case report forms. Patients with CAS underwent computed tomography angiography (CTA) or digital subtraction angiography (DSA), and RSSI patients underwent high-resolution vessel wall imaging to assist etiological classification.
Diagnostic and exclusion criteria
Group definitions were prespecified, based on established disease-specific criteria, and applied hierarchically to ensure mutually exclusive categories. CAS was defined as acute ischemic stroke attributed to atherosclerotic carotid artery stenosis > 70% or occlusion, assessed by CTA or DSA according to the North American Symptomatic Carotid Endarterectomy Trial (NASCET) method[
16]. RSSI was defined as a single acute infarct located in the territory of a penetrating artery, accompanied by corresponding neurological symptoms, without > 50% stenosis of the parent artery supplying the infarct territory[
1,
17]. No strict infarct diameter threshold was imposed; however, infarct location, morphology, vascular territory, parent-artery stenosis, and vessel wall imaging findings were jointly reviewed to reduce inclusion of other causes. Chronic CSVD was defined as periventricular white matter hyperintensity (PWMH) or deep white matter hyperintensity (DWMH) Fazekas score ≥ 2, in the absence of acute infarction and without a definite history of stroke or transient ischemic attack[
18]. Controls were defined as individuals with both PWMH and DWMH Fazekas scores of 0 or 1, without acute infarction, prior stroke or transient ischemic attack, or significant extracranial or intracranial arterial stenosis.
Acute infarction was defined as an ischemic lesion occurring within 14 days of symptom onset, accompanied by corresponding neurological deficits and characterized by hyperintensity on diffusion-weighted imaging (DWI) and hypointensity on the apparent diffusion coefficient map.
Patients in the CAS and RSSI groups were excluded if they had conditions that could confound stroke etiology classification, including atrial fibrillation, other major cardioembolic sources, Moyamoya disease, arterial dissection, radiation-induced carotid vasculopathy, vasculitis, or other non-atherosclerotic vasculopathies. Individuals with additional neurological disorders that could affect brain structure, including Alzheimer’s disease, Parkinson’s disease, brain tumor, demyelinating disease, or major traumatic brain injury, were also excluded. Patients with large territorial infarctions were not included because these lesions may substantially interfere with automated brain segmentation.
OCT/OCTA imaging
Retinal imaging was performed using a swept-source OCT/OCTA system (VG200S, version 2.0.106; SVision Imaging, Henan, China)[
13,
19], with a central wavelength of 1,050 nm and a scanning speed of 200,000 A-scans/s. Both eyes were scanned using a 6 × 6 mm
2 macular protocol consisting of 512 A-scans × 512 B-scans, with four repeated B-scans at each location. The same acquisition protocol, segmentation algorithm, and quality-control criteria were applied across all study groups.
Retinal layer segmentation and vascular measurements were generated automatically using the built-in software. Structural OCT parameters included retinal nerve fiber layer (RNFL) thickness and ganglion cell–inner plexiform layer (GCIPL) thickness. OCTA metrics were derived from the platform’s standard slab segmentation. The superficial vascular complex (SVC) extended from the internal limiting membrane to the inner two-thirds of the GCIPL, and the deep vascular complex (DVC) extended from the outer one-third of the GCIPL to the outer boundary of the outer plexiform layer. The choriocapillaris (CC) was defined as a 20-μm slab beneath the retinal pigment epithelium–Bruch’s membrane complex. Vessel length density (VLD, mm−1) was calculated as the total length of the skeletonized OCTA-positive vascular network per unit area. Because OCTA is based on motion contrast from moving erythrocytes, VLD is a flow-signal-dependent morphometric measure of the detectable vascular network and is not a direct measurement of volumetric blood flow or tissue perfusion.
Image quality was assessed according to the OSCAR-IB criteria (obvious problems, poor signal strength, centration of scan, algorithm failure, retinal pathology other than MS (multiple sclerosis)-related, illumination and beam placement)[
20] and APOSTEL (advised protocol for OCT study terminology and elements) recommendations[
21]. Scans with a signal quality score < 8, severe motion artifacts, segmentation failure, or obvious structural distortion were excluded. Eyes with ocular conditions that could affect OCT/OCTA acquisition or alter retinal structure were also excluded, including visually significant cataract, age-related macular degeneration, diabetic retinopathy, glaucoma, vitreomacular traction, retinal vascular occlusion, optic neuropathy, and pathological myopia. Participants were excluded only when neither eye met the predefined quality criteria; otherwise, all eligible eyes were retained for analysis.
MRI acquisition and analysis
All participants underwent brain MRI, including T1-weighted imaging, T2-weighted imaging, fluid-attenuated inversion recovery (FLAIR), diffusion-weighted imaging, and time-of-flight magnetic resonance angiography. T1-weighted and FLAIR images used in the present analysis were acquired with two-dimensional sequences on 3.0-T scanners. Detailed acquisition parameters are provided in the Supplementary Materials.
To accommodate variations in scanner type, acquisition parameters, and slice thickness, T1-weighted images were processed using SynthSeg version 2.0 in robust prediction mode[
22]. Intracranial volume (ICV) was obtained from the same pipeline. Regional brain volumes were extracted based on the Desikan–Killiany atlas[
23]. To reduce multiple-comparison burden and improve robustness in relatively low-resolution images, fine-grained anatomical labels were merged into larger regions of interest, as described in the supplementary methods. Global gray matter (GM), white matter (WM), and cerebrospinal fluid volumes were also calculated. All global and regional brain volumes were normalized to ICV.
WMHs were segmented from FLAIR images using WMH-SynthSeg[
24], which generates voxel-wise WMH probability maps. Binary WMH masks were obtained using a probability threshold of 0.1, consistent with our previous work[
14]. Based on the WMH masks and anatomical parcellations, total WMH volume was further divided into PWMH and DWMH. DWMH was defined as WMH located > 10 mm from the ventricular margins, whereas PWMH was defined as WMH within 10 mm of the ventricular margins. WMH volumes were normalized to ICV.
SynthSeg and WMH-SynthSeg outputs were resampled to 1.0 × 1.0 × 1.0 mm3 isotropic resolution and standardized to a common orientation before quantitative analysis. Segmentation quality was evaluated using both model-generated quality-control scores and visual inspection. A randomly selected 10% of T1-weighted segmentations was manually reviewed by an experienced researcher for anatomical plausibility, tissue-boundary accuracy, and major segmentation errors. Based on the correspondence between automated scores and manual assessments, scans with a white-matter segmentation quality control (QC) score < 0.75 or a gray-matter segmentation QC score < 0.70 were excluded. In total, 24 scans either failed processing or fell below these predefined thresholds. WMH masks in both acute stroke groups were visually reviewed together with the corresponding FLAIR, DWI, and apparent diffusion coefficient (ADC) images. The prespecified probability threshold of 0.1 was retained throughout the analysis. When FLAIR hyperintensity corresponding unequivocally to an acute infarct was included in the WMH mask, the misclassified voxels were manually removed before calculation of total WMH, PWMH, and DWMH volumes. Manual correction was performed conservatively and was limited to regions clearly corresponding to acute infarction on DWI and ADC. Correction was required in 36 participants with CAS and 4 participants with RSSI.
Statistical analysis
Continuous variables were assessed using quantile–quantile plots and are presented as mean ± standard deviation or median (interquartile range), as appropriate. Categorical variables are summarized as counts and percentages. Baseline characteristics were compared among groups using analysis of variance, the Kruskal–Wallis test, the χ2 test, or Fisher’s exact test, as appropriate.
Eye-level OCT/OCTA metrics were compared among groups using generalized estimating equations with an exchangeable working correlation structure to account for inter-eye correlation. Patient-level MRI-derived metrics were compared using multivariable linear regression models. Associations between OCT/OCTA and MRI-derived metrics were examined using generalized estimating equations with standardized variables. To test whether retina–brain associations differed across disease subtypes, interaction terms between MRI-derived metrics and disease group were included, with RSSI as the reference group. Group-specific slopes were estimated from the interaction models.
All multivariable models were adjusted for age, sex, hypertension, diabetes, dyslipidemia, smoking, and drinking. To evaluate the robustness of the findings to between-group covariate imbalance, we conducted sensitivity analyses using generalized propensity-score overlap weighting for the four study groups (supplementary methods). Propensity scores were estimated from age, sex, hypertension, diabetes, dyslipidemia, smoking, and drinking. Patient-level weights were applied to generalized estimating equations for eye-level outcomes and weighted linear models for patient-level MRI outcomes. Covariate balance was evaluated using absolute standardized mean differences, with values < 0.10 indicating adequate balance. We additionally repeated the principal analyses after excluding participants with diabetes. An exploratory side-specific analysis was conducted among CAS participants with unilateral qualifying carotid stenosis or occlusion and two eligible eyes. OCT/OCTA measurements in the eye ipsilateral to the stenotic carotid artery were compared with those in the contralateral eye using generalized estimating equations. Participants with bilateral qualifying stenosis or only one eligible eye were excluded. For pairwise group comparisons and association analyses, P values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method within each predefined analysis family. FDR-adjusted P values < 0.05 were considered statistically significant.
Results
Study population and baseline characteristics
A total of 811 individuals were initially screened. After applying the eligibility and image-quality control criteria, 651 participants with 1,257 eligible eyes were included in the final analysis (Figure 1). The final cohort comprised 228 controls with 434 eligible eyes, 283 patients with CAS with 551 eligible eyes, 74 patients with RSSI with 144 eligible eyes, and 66 patients with CSVD with 128 eligible eyes. Overall, 606 participants (93.1%) contributed two eligible eyes and 45 (6.9%) contributed one eligible eye.
Baseline characteristics are summarized in Table 1. The four groups differed significantly in age, hypertension, dyslipidemia, diabetes, smoking, and drinking status (all P < 0.05), whereas sex distribution was comparable across groups (P = 0.199). The CSVD group was older, and vascular risk profiles differed substantially across groups. Among patients with acute ischemic stroke, the median interval from symptom onset to OCT/OCTA examination was 8 [5, 10] days in the CAS group and 6 [5, 9] days in the RSSI group. The median National Institutes of Health Stroke Scale (NIHSS) score at OCT/OCTA examination was 3 [1, 3] in CAS and 3 [1, 4] in RSSI. The median absolute interval between OCT/OCTA and MRI examinations was 2 days in each study group.
Group differences in retinal and MRI-derived imaging metrics
Group differences in retinal OCT/OCTA metrics are shown in Figure 2 and Supplementary Table 1. After multivariable adjustment and FDR correction, the CAS group showed lower GCIPL thickness than both controls and RSSI (both FDR-adjusted P < 0.001). CAS also showed consistently lower SVC, DVC, and CC VLD than controls and RSSI (all FDR-adjusted P < 0.001). Compared with RSSI, the CSVD group had lower VLD in the SVC, DVC, and CC layers (FDR-adjusted P = 0.002, 0.004, and 0.004, respectively), although the magnitude of these differences was smaller than that observed for CAS.
For MRI-derived metrics, all patient groups showed greater total WMH, PWMH, and DWMH burden (all FDR-adjusted P < 0.001) compared with controls. CAS showed lower GM/ICV and WM/ICV than controls and RSSI (all FDR-adjusted P < 0.001). Among patient groups, RSSI showed greater total WMH, PWMH, and DWMH burden than CSVD (FDR-adjusted P < 0.001, < 0.001, and = 0.002, respectively). Given the potential contribution of infarct-related signal abnormalities to automated WMH segmentation in stroke patients, these findings were interpreted cautiously.
The generalized overlap-weighted analysis retained the principal CAS-related findings, including lower GCIPL thickness and lower SVC, DVC, and CC VLD compared with controls and RSSI (Supplementary Table 2 and Supplementary Figure 1). In an additional analysis restricted to participants without diabetes, these CAS-related differences also remained significant (all FDR-adjusted P < 0.001; Supplementary Table 3). The principal MRI findings were generally unchanged, including lower GM/ICV and WM/ICV in CAS and greater WMH burden in all patient groups than in controls. In contrast, secondary comparisons between RSSI and CSVD varied across the sensitivity analyses and were therefore interpreted cautiously.
Among 258 CAS participants included in the paired-eye analysis, ipsilateral eyes showed lower SVC and DVC VLD than contralateral eyes (FDR-adjusted P < 0.001 and 0.002, respectively; Supplementary Table 4).
Group-Specific associations between OCTA and global MRI markers
Patterns of OCTA–MRI associations varied across disease categories (Figure 3 and Supplementary Table 5). For GM/ICV, SVC VLD was positively associated with GM/ICV in the CAS group (standardized β = 0.238, 95% confidence interval [CI]: 0.164 to 0.311), and this association was stronger than the corresponding association in RSSI (FDR-adjusted P for interaction = 0.032). Positive associations with GM/ICV were also observed for DVC VLD and CC VLD in CAS (standardized β = 0.174, 95% CI 0.105 to 0.244; standardized β = 0.281, 95% CI 0.207 to 0.354, respectively), although the corresponding interaction terms did not remain significant.
For WMH-related metrics, inverse associations were observed primarily in the CAS group. In CAS, SVC VLD was inversely associated with total WMH/ICV (standardized β = −0.222, 95% CI: −0.305 to −0.138; FDR-adjusted P for interaction vs. RSSI = 0.019) and PWMH/ICV (standardized β = −0.242, 95% CI: −0.325 to −0.159; FDR-adjusted P for interaction vs. RSSI = 0.046). For DWMH/ICV, group differences in slopes were observed for SVC and DVC VLD, but the within-group associations were weaker and less stable. Overall, OCTA-derived VLD, particularly SVC VLD, showed the most consistent associations with GM volume and WMH-related measures in CAS, whereas corresponding associations in RSSI and CSVD were weaker, less consistent, or imprecisely estimated.
Regional brain volume associations
At the regional level, SVC VLD was associated with several cortical and subcortical brain volumes within the CAS group, whereas fewer associations reached statistical significance in RSSI and CSVD (Figure 4 and Supplementary Table 6). In CAS, higher SVC VLD was associated with larger thalamic volume (standardized β = 0.248, 95% CI 0.165 to 0.331), hippocampal volume (standardized β = 0.194, 95% CI 0.120 to 0.268), and multiple cortical regions, including the frontal, temporal, parietal, occipital, and cingulate regions.
Compared with RSSI, the association slopes in CAS differed significantly for the basal ganglia, hippocampus, temporal lobe, and cingulate region (FDR-adjusted P for interaction = 0.046, 0.034, 0.026, and 0.014, respectively). In RSSI, significant regional associations were limited to a positive association with basal ganglia volume and an inverse association with brainstem volume. In CSVD, regional association estimates were less precise and fewer associations reached statistical significance. Given the smaller sample size, these findings should not be interpreted as evidence of absent retina–brain associations.
Discussion
In this multicenter retinal and brain imaging study, we compared OCT/OCTA and quantitative MRI profiles across controls, CAS, RSSI, and chronic CSVD. Three main findings emerged. First, CAS showed the most pronounced OCTA-defined microvascular rarefaction, with lower VLD across the SVC, DVC, and CC layers, accompanied by reduced GCIPL thickness. Second, although CAS, RSSI, and CSVD all showed greater WMH burden than controls, their retinal imaging patterns differed, suggesting that OCTA alterations were unlikely to reflect WMH burden alone. Third, OCTA–MRI associations were most consistent in CAS, particularly for SVC VLD, which was associated with GM volume, WMH-related measures, and several regional brain volumes. The generalized overlap-weighted sensitivity analysis supported the principal CAS-related findings, including lower GCIPL thickness and lower VLD across all three vascular layers compared with controls and RSSI. In contrast, the differences in VLD between CSVD and RSSI were attenuated after weighting, suggesting that these secondary contrasts were sensitive to differences in age and vascular risk profiles. Together, these findings suggest that retinal OCTA may provide complementary subtype-relevant information across different cerebrovascular disease categories.
The pronounced reduction in OCTA-derived VLD in CAS is biologically plausible given the anatomical and hemodynamic links between the carotid and ocular circulations. The retina and choroid receive blood supply through the ophthalmic artery, and carotid stenosis may therefore influence ocular vascular networks through chronic atherosclerotic disease, impaired perfusion pressure, and territory-related hemodynamic stress[
9,
19,
25,
26]. Previous OCTA studies[
9,
13] have reported reduced retinal vascular density in carotid stenosis, ipsilateral retinal microvascular alterations, associations with plaque characteristics or stenosis severity, and short-term vascular changes after carotid revascularization. These observations are consistent with our finding of broad VLD reduction in CAS. However, OCTA-derived VLD should be interpreted as a measure of the visible retinal microvascular network rather than a direct measure of tissue perfusion[
8]. Moreover, although embolic and ischemic ocular manifestations have been described in carotid disease[
19,
27], lesion-specific retinal ischemic signs were not the focus of the present study. An exploratory paired-eye analysis additionally showed lower SVC and DVC VLD in eyes ipsilateral to the qualifying carotid lesion, supporting a side-specific retinal microvascular component in CAS. However, the contralateral eye should not be regarded as a healthy control because both eyes share systemic vascular risk factors, and this secondary analysis should be interpreted cautiously.
In contrast, RSSI and chronic CSVD did not show the same consistently detected global OCTA VLD reduction as CAS despite greater WMH burden. Although the primary analysis suggested lower VLD in CSVD than in RSSI, these differences were not retained after generalized overlap weighting and should therefore not be interpreted as robust subtype-specific differences. This pattern should not be interpreted as evidence of absent retinal involvement. RSSI represents an acute small subcortical infarct within the broader spectrum of small-vessel disease, whereas chronic CSVD reflects accumulated and heterogeneous small-vessel brain injury[
1]. Compared with CAS, these conditions may involve more localized perforator disease, diffuse arteriolosclerosis, endothelial dysfunction, blood-brain barrier disruption, and age- or hypertension-related white matter injury. Such mechanisms may not produce the same global retinal OCTA pattern as large-artery stenosis[
1,
28]. Although clearly misclassified acute infarct voxels were manually removed in both acute stroke groups, minor residual overestimation of WMH burden cannot be excluded, particularly in patients with small scattered or watershed white-matter infarcts. The smaller RSSI and CSVD groups limited the precision of subgroup-specific estimates and interaction tests. Accordingly, the absence of statistically significant associations in these groups should not be interpreted as evidence that retinal abnormalities or retina–brain relationships are absent.
Previous studies have linked retinal microvascular abnormalities with WMH burden and composite CSVD markers, although findings vary across cohorts[
15,
29]. Differences in CSVD definition, lesion burden, vascular risk profiles, and the use of global rather than lesion-specific retinal metrics may partly explain the less consistent associations observed in the present study.
The OCTA–MRI association analyses provide an additional perspective. In our study, lower OCTA-derived VLD was most consistently associated with GM volume and WMH-related measures in CAS. Regional analyses further showed associations with both subcortical and cortical brain volumes, including the thalamus, hippocampus, and several cortical regions. These findings suggest that, in CAS, retinal microvascular rarefaction may parallel broader MRI-defined brain tissue injury. Nevertheless, these associations should be interpreted cautiously. OCTA and MRI capture different biological compartments: OCTA measures the visible retinal microvascular network, whereas MRI-derived WMH and brain volumes reflect downstream cerebral tissue injury. Therefore, the stronger OCTA–MRI associations in CAS are more likely to indicate shared vulnerability of retinal and cerebral tissues under large-artery atherosclerotic disease than a direct causal pathway from retinal vascular loss to brain atrophy.
RNFL thickness did not show robust between-group differences. The small CAS–RSSI difference observed in the primary analysis was attenuated after overlap weighting, whereas the CAS-related differences in GCIPL thickness and OCTA-derived VLD remained. This pattern suggests that macular ganglion-cell and microvascular measures may be more sensitive than RNFL thickness to the retinal alterations detected in the present CAS cohort.
This study has several strengths. We directly compared CAS, RSSI, chronic CSVD, and controls within a unified OCT/OCTA and MRI framework. Retinal imaging was analyzed at the eye level while accounting for inter-eye correlation, and quantitative MRI-derived brain volumes and WMH measures were incorporated to assess both retinal features and retina–brain associations. In addition, interaction models were used to test whether OCTA–MRI associations differed across disease categories, rather than assuming a uniform retina–brain relationship across all cerebrovascular conditions.
Several limitations should also be acknowledged. First, the cross-sectional design precludes causal inference and does not allow assessment of temporal changes after stroke or carotid revascularization. Second, the groups differed substantially in age and vascular risk profiles. Although the principal CAS-related findings remained after generalized overlap weighting and measured covariates achieved improved balance, residual confounding from unmeasured or imperfectly measured factors cannot be excluded. In addition, overlap-weighted estimates pertain primarily to participants with covariate profiles shared across the four groups and may not be directly generalizable to individuals with more extreme disease-specific risk profiles. Third, OCTA-derived VLD reflects visible vascular density rather than direct blood flow or tissue perfusion and may be influenced by image quality, ocular magnification, and subclinical ocular conditions. Fourth, the chronic CSVD group was defined mainly by WMH burden and may not capture the full spectrum of CSVD imaging markers. Fifth, in the acute stroke groups, infarct-related FLAIR hyperintensity could occasionally be misclassified as WMH, particularly in CAS patients with small scattered or watershed white-matter infarcts. All WMH masks were reviewed with reference to DWI and ADC images, and clearly misclassified infarct regions were manually removed. Nevertheless, complete separation of small acute white-matter infarcts from pre-existing DWMH was not always possible on the relatively low-resolution two-dimensional FLAIR images. Minor residual overestimation of WMH burden in CAS and RSSI therefore cannot be excluded. Finally, MRI data were acquired in a real-world multicenter setting with heterogeneous protocols. Although robust segmentation methods and quality-control procedures were applied, segmentation errors and scanner-related variability remain possible.
Conclusion
In conclusion, CAS showed the most prominent OCTA-defined retinal microvascular rarefaction and the most consistent associations between retinal VLD and MRI-derived brain injury. RSSI and chronic CSVD also showed greater WMH burden than controls, whereas their OCTA differences and OCTA–MRI associations were less consistently detected and were estimated with lower precision. These findings suggest that retinal OCTA may provide complementary information for characterizing cerebrovascular disease categories, particularly in CAS. Longitudinal studies with external validation, standardized MRI protocols, and post-revascularization follow-up are needed to determine the temporal stability, reversibility, and clinical utility of these retinal imaging markers.
PDF
(1890KB)
