Multimodal Retinal Imaging Correlates With Brain Atrophy and Plasma Biomarkers in Alzheimer's Disease

Abstract

Purpose

To investigate retinal structural and microvascular changes in Alzheimer's disease (AD) and evaluate their associations with neuroimaging markers, plasma biomarkers, and cognitive impairment.

Methods

This cross-sectional study included 181 participants (95 with AD and 86 cognitively unimpaired controls). AD diagnosis was established according to National Institute on Aging–Alzheimer's Association criteria with positive amyloid-β positron emission tomography imaging. Retinal structure and microvasculature were assessed using swept-source optical coherence tomography (OCT) and OCT angiography (OCTA). Cortical thickness and gray matter volume were quantified using structural magnetic resonance imaging. Plasma Aβ42 and phosphorylated tau-217 (p-tau217) were measured, and cognitive function was evaluated using the Mini-Mental State Examination and Montreal Cognitive Assessment (MoCA).

Results

Compared with controls, patients with AD demonstrated reduced retinal nerve fiber layer, ganglion cell–inner plexiform layer (GCIPL), inner nuclear layer, and outer retinal layer thicknesses, together with lower superficial vascular complex (SVC) and deep vascular complex densities (all P < 0.001). GCIPL thickness and SVC density were associated with cortical atrophy in AD-vulnerable regions, including temporal, parietal, and occipital cortices. Lower plasma Aβ42 levels were associated with reduced retinal layer thickness and vascular density, whereas higher p-tau217 levels were associated with reduced outer retinal layer thickness and SVC density. GCIPL thickness showed the strongest association with cognitive performance, while lower SVC density was associated with poorer MoCA scores. APOE ε4 carriers exhibited greater GCIPL thinning, reduced retinal vascular density, and more extensive cortical abnormalities despite similar plasma Aβ42 and p-tau217 levels.

Conclusions

Retinal neuronal and microvascular alterations are associated with AD-related molecular pathology, cerebral neurodegeneration, APOE ε4-related vulnerability, and cognitive impairment. OCT/OCTA may provide a noninvasive adjunct approach for characterizing neurodegenerative and neurovascular changes in AD.

Translational Relevance

Retinal imaging may provide a scalable and noninvasive approach for detecting and monitoring AD pathology.

Introduction

Alzheimer's disease (AD) is the leading cause of dementia worldwide and is characterized by progressive cognitive deterioration accompanied by amyloid-β accumulation, tau hyperphosphorylation, synaptic dysfunction, and neuronal loss.^1,2 Pathologic alterations begin years before overt clinical symptoms emerge, highlighting the importance of identifying biomarkers capable of detecting early disease-related changes.³ Although cerebrospinal fluid (CSF) biomarkers and molecular neuroimaging have substantially improved the biological characterization of AD, their clinical implementation remains limited by cost, invasiveness, and accessibility. Consequently, there is increasing interest in identifying noninvasive and scalable biomarkers that can capture both neurodegenerative and vascular mechanisms associated with AD.

The current biological framework for AD relies on biomarkers reflecting amyloid deposition, tau pathology, and neurodegeneration. CSF biomarkers and neuroimaging techniques such as positron emission tomography (PET) and magnetic resonance imaging (MRI) have substantially improved the detection and characterization of AD pathology.^2,4 However, these approaches are limited by cost, invasiveness, and limited availability in routine clinical practice. Blood-based biomarkers, including plasma Aβ and phosphorylated tau species, have recently emerged as promising minimally invasive alternatives for detecting AD pathology.^5–7 Despite these advances, additional noninvasive and scalable biomarkers that capture structural and vascular manifestations of neurodegeneration are still needed.

The retina has emerged as a promising surrogate tissue for studying neurodegenerative disorders because of its developmental and anatomic relationship with the central nervous system. Retinal neurons and microvascular networks share structural and physiologic similarities with cerebral tissue, including tightly regulated neurovascular interactions and barrier systems.^8–10 These anatomic and physiologic similarities suggest that pathologic processes occurring in the brain may also be reflected in the retina.

Experimental and histopathologic studies support this hypothesis. Amyloid-β deposits, tau pathology, neuroinflammation, and degeneration of retinal ganglion cells have been identified in retinal tissues of individuals with AD.^11–13 In addition to neuronal degeneration, vascular dysfunction has emerged as an important contributor to AD pathophysiology.^14–16 Disruption of the neurovascular unit—including endothelial cells, pericytes, astrocytes, and neurons—can impair cerebral perfusion and vascular integrity. Because the retinal vasculature shares many structural and functional features with the cerebral microvasculature, retinal vascular alterations may provide a surrogate marker of neurovascular dysfunction in AD.

Recent advances in retinal imaging technologies allow detailed in vivo assessment of retinal neuronal integrity and vascular perfusion. Optical coherence tomography (OCT) provides quantitative measurements of intraretinal layer thickness, whereas optical coherence tomography angiography (OCTA) enables noninvasive visualization of retinal capillary networks. Previous studies have reported retinal thinning and impaired retinal perfusion in individuals with AD, suggesting that retinal imaging may capture both neurodegenerative and microvascular alterations related to disease progression.^13,17–20

However, the relationship between retinal imaging markers, cerebral neurodegeneration, and emerging blood-based biomarkers remains incompletely understood. In addition, the influence of genetic risk factors such as the apolipoprotein E (APOE) ε4 allele on retinal alterations in AD has not been fully clarified. In this study, we investigated retinal structural and microvascular changes in individuals with AD using OCT and OCTA and examined their associations with cortical thickness and gray matter volume measured by MRI, plasma biomarkers of AD pathology, and global cognitive performance. We further explored whether these retinal alterations differed according to APOE ε4 carrier status, intending to clarify the relationship between retinal changes and established markers of AD pathology.

Methods

Supplementary Figure S1 shows the study's central concept. This cross-sectional observational study was conducted at the Geriatrics Department of the First Affiliated Hospital of Wenzhou Medical University. As part of our research protocol, all participants underwent retinal scans via retinal imaging via OCT/OCTA, 3.0T brain MRI scans, and neuropsychological evaluations. Demographic data such as sex, age, education, and vascular risk factors (hypertension and diabetes mellitus) were recorded.

All participants completed cognitive screening using the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA). Patients with AD were required to meet the National Institute on Aging–Alzheimer's Association diagnostic criteria.⁴ The inclusion criteria for the AD group were (1) presence of clinical dementia with a positive amyloid-β PET scan and an MMSE score below 26, (2) availability of complete and detailed clinical data, (3) normal optic disc and retinal morphology on ophthalmic examination, and (4) ability to cooperate with and complete MRI and OCT/OCTA examinations.

The exclusion criteria for patients with AD were (1) diagnosis of other dementia subtypes, including dementia with Lewy bodies, frontotemporal dementia, or vascular dementia; (2) current or prior ocular conditions that could influence retinal measurements, such as diabetic retinopathy, optic neuropathy, or advanced cataract; (3) history or presence of other neurologic or psychiatric disorders, including stroke or schizophrenia; (4) history of vascular diseases, including poorly controlled hypertension or diabetes mellitus; and (5) history of malignancy.

Cognitively unimpaired (CU) participants were recruited through public advertisements and consisted of individuals living near the hospital or undergoing routine annual health examinations at our institution. Participants with no self-reported history of neurologic disease, no abnormalities on brain MRI, and normal cognitive performance were enrolled as controls. Demographic and clinical information were collected for all CU participants, and the exclusion criteria were consistent with those applied to the AD group.

Plasma Analyses

Venous blood samples were obtained during morning hours; however, participants were not required to fast prior to collection. Blood was drawn into six K2-EDTA anticoagulant tubes and delivered to the laboratory within 3 hours for processing. Samples were centrifuged at 2000 × g at 4°C for 10 minutes. The resulting plasma was carefully separated, aliquoted into 1.5-mL polypropylene cryovials (1 mL per aliquot), and stored at −80°C until analysis. Plasma collection occurred within 1 month of the corresponding neuroimaging, retinal imaging, and clinical data assessments. All biochemical analyses were conducted at Nanjing Vazyme Biotech (Nanjing, China), as previously reported.^1,21

On the day of measurement, plasma aliquots were thawed at room temperature. To prevent variability related to repeated freeze–thaw cycles, only previously unthawed aliquots were used. Concentrations of amyloid-β42 (Aβ42) and phosphorylated tau-217 (p-tau217) were quantified from 0.5-mL plasma samples using a fully automated chemiluminescent immunoassay system (Elf S240; Vazyme, Nanjing, China). The corresponding commercial assay kits (Vazyme β-Amyloid 1–42 and Vazyme pTau217 Plasma) were applied according to the manufacturer's standardized operating procedures. Following thawing, samples were vortex-mixed and centrifuged to remove potential fibrin debris prior to assay loading.¹

All participants underwent standardized 3.0-T MRI (SIGNA MR; GE Healthcare, Waukesha, WI, USA) using a uniform acquisition protocol. High-resolution T1-weighted structural images were obtained with a magnetization-prepared rapid gradient-echo (MPRAGE) sequence and subsequently processed for cortical reconstruction and volumetric analyses as previously reported.^1,22,23 For the present analysis, high-resolution T1-weighted images were obtained with a MPRAGE sequence using the following parameters: repetition time (TR) of 2000 ms, echo time (TE) of 2.19 ms, inversion time of 900 ms, field of view of 240 × 240 mm², acquisition matrix of 256 × 256, flip angle of 8°, and an isotropic voxel size of approximately 0.9 × 0.94 × 0.94 mm³.

Quantification of Cortical Thickness and Gray Matter Volume

Regional measures of cortical thickness and gray matter volume were obtained using FreeSurfer (version 7-dev; http://surfer.nmr.mgh.harvard.edu/) via the “recon-all” surface reconstruction pipeline. Briefly, individual three-dimensional (3D) T1-weighted images were used as input for cortical reconstruction. To reduce potential misclassification of non–brain tissues as cortex, brain masks generated by SynthStrip were applied in place of the default FreeSurfer skull-stripping procedure.²⁴ Cortical thickness and gray matter volume were then extracted for each region defined by the Desikan–Killiany cortical parcellation, comprising 34 regions per hemisphere, and were used as a regional index of cortical atrophy.²⁵ Total cortical thickness and gray matter volume were determined by summing the volumes in the left and right hemispheres. A general linear model was applied to compare the cortical thickness between AD and CU while adjusting for covariates such as age, sex, hypertension, and educational years. Clusters were defined using a vertex-wise threshold of P < 0.001, followed by cluster-level correction for multiple comparisons using random field theory.²⁶

Swept-Source OCTA Imaging

Retinal microvasculature was acquired by a swept-source optical coherence tomography angiography system (SS-OCTA, VG 200; SVision Imaging, Henan, China; version 3.1.352) between 8 AM and 4 PM. The SS-OCTA contained a swept-source light source with a central wavelength of 1050 nm and a scan rate of 200,000 A-scans per second, as previously reported.^{17,20,27–32} The tool was also equipped with an eye-tracking utility based on an integrated confocal scanning laser ophthalmoscope to reduce and eliminate eye motion artifacts. The axial resolution, lateral resolution, and scan depth were 5 µm, 13 µm, and 3 mm, respectively. Imaging was done in both left and right eyes without pupil dilation.

After imaging, data derived from the images were processed automatically by an algorithm within the OCT tool to segment the intraretinal layers as described in previous reports. In the current study, the macula-centered retinal structure was segmented into retinal nerve fiber layer (RNFL), ganglion cell–inner plexiform layer (GCIPL), inner nuclear layer (INL), and outer retinal layer (ORL). For each eye included in this study, a 3D thickness map of the intraretinal layers in each eye was generated based on the segmented layers. The mean structural thickness, measured in micrometers, was used in our study.

OCTA images were acquired using a raster scanning protocol consisting of 512 horizontal B-scans over a 6 × 6–mm² region centered on the fovea. Each B-scan included 512 A-scans and was repeated eight times for averaging. Only angiograms with a signal quality score of 8 or higher were included in the analysis. En face angiograms of the superficial vascular complex (SVC) and deep vascular complex (DVC) were generated using automated layer segmentation to assess retinal perfusion. The boundary between the SVC and DVC was defined at the interface between the inner two-thirds and outer one-third of the GCIPL. Retinal microvascular status was quantified using vessel density (%). The reliability of the OCTA parameters applied in this study has been validated previously.³³ Image quality control adhered to the OSCAR-IB quality criteria³⁴ and the APOSTEL recommendation.³⁵ All angiograms used for processing followed the OSCAR-MP Consensus Criteria for Quality Assessment of OCTA.³⁶

Images affected by ocular conditions that could compromise retinal measurements, including moderate-to-severe cataract, optic neuritis, or age-related macular degeneration, were excluded. When only one eye was affected, data from the contralateral eye were analyzed; participants with bilateral pathology were excluded entirely.

Ethics

The study was approved by the Ethics Committee of the First Affiliated Hospital of Wenzhou Medical University and followed the Declaration of Helsinki. We obtained signed informed consent from all participants or their legal guardians before enrolling in our study. We followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline.

Statistical Analyses

Continuous variables with approximately normal distributions are presented as mean ± standard deviation, whereas nonnormally distributed variables are summarized using medians and interquartile ranges. Categorical data are reported as counts and percentages. Differences in baseline characteristics between groups were evaluated using the Wilcoxon rank-sum test for continuous variables and the Pearson χ² test for categorical variables.

Comparisons of OCT and OCTA parameters between CU and AD, as well as APOE ε4 carriers versus noncarriers in AD, were performed using multiple linear regression models with generalized estimating equations (GEEs) to account for intereye correlations. Models were adjusted for age, sex, hypertension, diabetes mellitus, years of education, and intereye dependency, with an exchangeable working correlation structure.

Comparison of neuroimaging measures between CU and AD was performed using linear regression models to account for total intracranial volume, age, sex, and educational years.

A multiple linear regression model with GEE was used to assess the association between OCT/OCTA-derived measures and neuroimaging metrics (cortical thickness and gray matter volume) in AD while adjusting for risk factors (age, sex, hypertension, diabetes mellitus, and educational years), intracranial volume, and intereye dependencies. Similarly, a multiple linear regression model with GEE was used to explore the association between OCT/OCTA-derived measures and plasma biomarkers in AD while adjusting for covariates. Results were reported as regression coefficients (β, per standard deviation [SD] increase/decrease) with corresponding 95% confidence intervals (CIs). Model estimates for the continuous predictors are presented as standardized betas (β, per SD increase/decrease). Analyses were also conducted separately for the left (oculus sinister) and right (oculus dexter) eyes for transparency.

Partial Spearman correlation was used to analyze the association between neuroimaging measures and blood plasma biomarkers, as well as other core AD characteristics, while adjusting for age, sex, hypertension, diabetes mellitus, educational years, and intracranial volume. The core AD measures included symptom severity (global Clinical Dementia Rating Scale [CDR]) and cognitive performance.

A multiple linear regression model with GEE was also used to investigate the association between OCT/OCTA-derived measures with symptom severity and global cognitive measures in AD while adjusting for age, sex, hypertension, diabetes mellitus, educational years, and intereye dependencies. Results were reported as regression coefficients (β) with corresponding 95% CIs.

We further tested the interaction between OCT/OCTA-derived measures, neuroimaging measures, and plasma biomarkers on global cognitive performance by including the cross‐product term of OCT/OCTA × neuroimaging measures or plasma biomarkers along with the main effect terms of each variable in the models. The same covariates used in previous analyses examining the relationship between neuroimaging measures and global cognitive performance were controlled.

All statistical analyses were conducted in RStudio using R version 4.4.4 (R Project for Statistical Computing, Vienna, Austria) with packages dplyr and gee. Plots were created with “ggplot2” and “ggcorrplot.” P values less than 0.05 were considered statistically significant.

Results

Demographics and Baseline Characteristics

A total of 202 individuals were initially enrolled. Twenty-one participants were excluded due to incomplete neuroimaging data, inability to complete study procedures, or retinal abnormalities identified during image evaluation (e.g., severe cataract or age-related macular degeneration). The final analytic sample comprised 181 participants, including 95 individuals with AD and 86 CU controls (Table 1).

Table 1.

Demographics and Clinical Information of Study Participants

Characteristic	CU (n = 86)	AD (n = 95)	P Value
Demographics
Age, y	65.98 ± 5.79	67.64 ± 8.14	0.118
Males	30 (34.88)	32 (33.68)	0.990
Duration, y		2 [1, 4]	0.314
APOE ε4 allele		35 (36.84)
Vascular risk factors
Hypertension	31 (36.05)	31 (32.63)	0.744
Diabetes mellitus	12 (13.95)	13 (13.68)	1
Hyperlipidemia	26 (30.23)	31 (32.63)	0.852
Global cognitive examinations
MMSE	29 [28, 30]	19 [14, 22]	<0.001
MoCA	28 [27, 30]	13 [10, 16]	<0.001
CDR	0	1 [0.5, 1]	<0.001
OCT/OCTA-derived measuresa
SVC, %	50.75 ± 4.34	39.74 ± 5.08	<0.001
DVC, %	55.27 ± 3.34	49.14 ± 6.76	<0.001
RNFL, µm	38.96 ± 3.21	35.43 ± 4.76	<0.001
GCIPL, µm	70.88 ± 3.88	60.03 ± 8.35	<0.001
INL, µm	39.97 ± 2.22	38.12 ± 2.20	<0.001
ORL, µm	153.75 ± 7.92	147.15 ± 9.64	<0.001
Plasma biomarkers
Aβ42, pg/mL	8.77 ± 1.58	6.65 ± 1.56	<0.001
p-tau 217, pg/mL	1.81 ± 0.95	8.99 ± 2.99	<0.001
p-tau 217/Aβ42	0.22 ± 0.12	1.47 ± 0.67	<0.001

Values are presented as number (%), mean ± standard deviation, or median [interquartile range]. Demographic characteristics and baseline clinical variables were compared using unadjusted Wilcoxon rank-sum tests or Pearson χ² tests.

^aA multiple linear regression model with a GEE was used to compare the parameters between CU and AD while adjusting for risk factors (age, sex, hypertension, diabetes mellitus, and education) and intereye dependencies.

Baseline demographic characteristics and vascular risk factors were comparable between groups (all P > 0.05). As expected, global cognitive performance was significantly lower in the AD group than in CU participants (MMSE and MoCA, both P < 0.001). Plasma biomarker profiles were also consistent with AD pathology, with lower plasma Aβ42 concentrations and higher plasma p-tau217 levels in AD than in CU (both P < 0.001).

Retinal Structural and Microvascular Differences Between AD and CU

Compared with CU participants, patients with AD exhibited significant reductions in retinal microvascular density in both the SVC and DVC (P < 0.001 for both comparisons).

Structural OCT analysis demonstrated widespread retinal thinning in AD, including significant reductions in RNFL, GCIPL, INL, and ORL thicknesses (all P < 0.001), indicating that retinal abnormalities extended beyond the ganglion cell complex.

Neuroimaging Differences Between AD and CU

Surface-based MRI analyses demonstrated widespread cortical atrophy in patients with AD compared with CU participants. The most pronounced cortical thinning involved regions characteristically affected by AD, including the superior, middle, and inferior temporal gyri; inferior parietal cortex; superior and middle occipital cortices; and cuneus (all cluster-corrected P < 0.05).

Gray matter volume analyses demonstrated a similar regional distribution, with significant volume loss overlapping areas of cortical thinning (Fig. 1).

Figure 1.

(A) Maps showing the spatial distribution of effect sizes for differences in cortical thickness and gray matter volume between individuals with AD and CU individuals. (B) Maps depicting the spatial distribution of effect sizes for cortical thickness and gray matter volume differences between APOE ε4 carriers and noncarriers in AD. Surface-based representations display Cohen's d effect sizes for both cortical thickness (top row) and gray matter volume (bottom row) comparisons across the two groups. Regions with lower values are shown in cooler colors. Both lateral and medial views of the left and right hemispheres are presented. The accompanying color bar indicates the magnitude and direction of the effect size.

Retinal and Neuroimaging Differences According to APOE ε4 Status in AD

Among the 95 participants with AD, 35 were APOE ε4 carriers, and 60 were noncarriers. Demographic characteristics and vascular risk factors did not differ significantly between groups (Table 2). However, global cognitive performance was lower among APOE ε4 carriers (P < 0.010).

Table 2.

Demographics and Clinical Information on AD Based on APOE ε4 Status

Characteristic	Noncarriers (n = 60)	Carriers (n = 35)	P Value
Demographics
Age, y	66.42 ± 8.34	69.74 ± 7.43	0.054
Males	21 (35.0)	11 (31.43)	0.896
Duration, y	3 [2, 4]	2 [1, 4]	0.314
Vascular risk factors
Hypertension	17 (28.33)	14 (40.0)	0.346
Diabetes mellitus	9 (15.0)	4 (11.43)	0.858
Hyperlipidemia	19 (31.67)	12 (34.29)	0.971
Global cognitive examinations
MMSE	20 [17, 22]	15 [12, 20]	0.002
MoCA	14 [12, 17]	11 [8, 13]	0.001
CDR	0.5 [0.5, 1]	1 [0.5, 1]	0.294
OCT/OCTA-derived measuresa
SVC, %	40.98 ± 4.30	37.26 ± 5.62	<0.001
DVC, %	50.48 ± 4.65	46.47 ± 9.14	0.016
RNFL, µm	35.72 ± 4.74	34.91 ± 4.78	0.493
GCIPL, µm	62.74 ± 7.34	55.31 ± 7.92	<0.001
INL, µm	38.45 ± 2.28	37.55 ± 1.96	0.073
ORL, µm	149.47 ± 9.54	145.36 ± 9.32	0.096
Plasma biomarkers
Aβ42, pg/mL	6.81 ± 1.56	6.45 ± 1.57	0.402
p-tau 217, pg/mL	8.99 ± 3.19	8.98 ± 2.77	0.986
p-tau 217/Aβ42	1.45 ± 0.72	1.49 ± 0.61	0.835

Values are presented as number (%), mean ± standard deviation, or median [interquartile range]. Demographic characteristics and baseline clinical variables were compared using unadjusted Wilcoxon rank-sum tests or Pearson χ² tests.

^aA multiple linear regression model with a GEE was used to compare the parameters between APOE ε4 in AD while adjusting for risk factors (age, sex, hypertension, diabetes mellitus, and education) and intereye dependencies.

APOE ε4 carriers exhibited more pronounced retinal microvascular impairment, with lower SVC (P < 0.001) and DVC (P = 0.016) densities compared with noncarriers, respectively. In addition, GCIPL thickness was significantly reduced in carriers (P < 0.001), whereas no significant differences were observed for RNFL, INL, or ORL thicknesses.

Neuroimaging analyses indicated greater cortical thinning among APOE ε4 carriers, particularly in the temporal, occipital, and inferior parietal cortices (all P < 0.05). Gray matter volume analyses showed a comparable regional pattern of atrophy (Fig. 1).

Associations Between Retinal Measures and Cerebral Neuroimaging in AD

Regression analyses demonstrated significant associations between retinal measures and regional cerebral atrophy in AD after adjustment for covariates. Reduced GCIPL thickness and lower SVC density were consistently associated with cortical thinning across several AD-vulnerable regions, including the supramarginal gyrus, superior and inferior parietal cortices, precuneus, lateral occipital cortex, pericalcarine cortex, lingual gyrus, precentral gyrus, caudal middle frontal cortex, and banks of the superior temporal sulcus (all P < 0.05; Fig. 2).

Figure 2.

Association between neuroimaging measures with GCIPL thickness and SVC density in AD. (A) Association between gray matter volume (GMV) with GCIPL and SVC in AD. (B) Association between cortical thickness with GCIPL and SVC in AD. OU (oculus uterque: both eyes), OD (oculus dexter: right eye), and OS (oculus sinister: left eye).

Reduced RNFL thickness was additionally associated with cortical thinning within the superior parietal and pericalcarine cortices. Similar regional associations were observed between retinal measures and gray matter volume (Supplementary Table S1).

Associations Between Retinal Measures and Global Cognitive Performance in AD

Retinal structural abnormalities were significantly associated with global cognitive impairment.

Among all retinal parameters, GCIPL thickness demonstrated the strongest and most consistent association with cognitive performance, showing significant positive associations with both MMSE and MoCA scores (both P < 0.001). Reduced INL thickness was likewise associated with lower MMSE and MoCA scores (both P < 0.05), whereas reduced ORL thickness was associated only with lower MMSE scores.

Regarding retinal microvasculature, lower SVC density was significantly associated with poorer MoCA performance (P = 0.037), whereas no significant association was observed with MMSE. No significant cognitive associations were identified for DVC density.

Associations Between Neuroimaging Measures and Global Cognition in AD

Lower MMSE scores were associated with reduced cortical thickness in the lateral orbitofrontal cortex, medial orbitofrontal cortex, and pars opercularis (all P < 0.05). No significant association was observed between gray matter volume and MMSE.

Lower MoCA scores were associated with reduced cortical thickness involving the lateral orbitofrontal, lingual, middle temporal, pars opercularis, and supramarginal cortices, together with lower gray matter volume in the middle temporal gyrus (all P < 0.05).

Associations Between Plasma Biomarkers and Cognitive Performance in AD

Lower plasma Aβ42, higher plasma p-tau217, and higher plasma p-tau217/Aβ42 were all significantly associated with poorer MoCA performance (all P ≤ 0.011).

Similarly, lower MMSE scores were associated with elevated plasma p-tau217 and higher plasma p-tau217/Aβ42 (both P < 0.05).

Associations of Retinal and Plasma Biomarkers in AD

Retinal structural and vascular alterations were significantly associated with circulating biomarkers of AD pathology.

Lower plasma Aβ42 levels were associated with reduced thicknesses of RNFL, GCIPL, INL, and ORL (all P < 0.05). In contrast, elevated plasma p-tau217 concentrations were significantly associated with reduced ORL thickness (P = 0.033).

Regarding retinal microvasculature, lower plasma Aβ42 levels were associated with reduced vessel density in both the SVC and DVC, whereas higher plasma p-tau217 concentrations were specifically associated with lower SVC density (P = 0.038) (Fig. 3).

Figure 3.

Correlation matrix of retinal microvascular, structural, cognitive, and AD-related plasma biomarkers. This correlogram illustrates the Pearson correlation coefficients (r) and associated significance levels for the multimodal parameters assessed in patients with AD. The color scale indicates the strength and direction of the correlation, ranging from –1 (red, negative) to +1 (blue, positive). Asterisks indicate statistical significance: *P < 0.05 and **P < 0.01.

Association Between Plasma Biomarkers and Cerebral Neurodegeneration

Lower plasma Aβ42 levels were associated with reduced gray matter volume within the inferior parietal cortex and banks of the superior temporal sulcus, together with cortical thinning involving the parahippocampal, inferior parietal, pericalcarine, and superior temporal cortices (all P < 0.05).

Higher plasma p-tau217 levels were associated with reduced gray matter volume in the lateral orbitofrontal, lingual, and superior temporal cortices and with cortical thinning involving the insula, isthmus cingulate, pars triangularis, and supramarginal gyrus (all P < 0.05) (Supplementary Table S2).

Associations With Disease Severity in AD

Greater clinical disease severity, reflected by higher global CDR scores, was significantly associated with thinner GCIPL, INL, and ORL, as well as lower SVC and DVC vessel densities (all P < 0.05).

Higher CDR scores were likewise associated with reduced cortical thickness and gray matter volume involving the caudal middle frontal, inferior parietal, middle temporal, precuneus, superior parietal, and supramarginal cortices (all P < 0.05) (Fig. 3 and Supplementary Table 3).

Interaction Analyses

Interaction analyses demonstrated a significant interaction between SVC density and middle temporal cortical thickness on global cognitive performance (β = 0.128; 95% CI, 0.009–0.247; P = 0.035), indicating that the relationship between retinal microvascular integrity and cognition varied according to the degree of cortical neurodegeneration.

A second significant interaction was observed between GCIPL thickness and the plasma p-tau217/Aβ42 ratio (β = −0.100; 95% CI, −0.193 to −0.007; P = 0.037), suggesting that retinal neuronal degeneration and systemic AD pathology jointly influenced cognitive performance.

No additional statistically significant interactions were identified.

Discussion

In this multimodal study integrating retinal OCT/OCTA, structural MRI, plasma biomarkers, and cognitive assessments, we demonstrated that individuals with biomarker-confirmed AD exhibited significant retinal neuronal and microvascular abnormalities compared with CU. Importantly, retinal structural and vascular alterations were closely associated with cortical atrophy within brain regions preferentially affected by AD, plasma biomarkers of amyloid and tau pathology, and global cognitive impairment. Furthermore, APOE ε4 carriers exhibited more pronounced retinal structural thinning and microvascular impairment and cortical degeneration despite comparable plasma biomarkers. Collectively, these findings support the concept that retinal imaging reflects multiple biological processes occurring during AD and may serve as an accessible, noninvasive complement to established biomarkers for evaluating neurodegeneration and microvascular dysfunction.

Our findings confirm and extend previous OCT studies demonstrating thinning of the RNFL and GCIPL in AD.^{11–13,18–20,37} These retinal layers contain retinal ganglion cell bodies and axons that form the optic nerve and are therefore particularly susceptible to neurodegenerative processes affecting the central nervous system.^38–40 Beyond these established findings, we also observed significant thinning of the INL and ORL, suggesting that retinal involvement in AD extends beyond the ganglion cell complex. However, the mechanisms underlying deeper retinal layer thinning are likely multifactorial and should be interpreted cautiously. Histopathologic studies have demonstrated retinal amyloid-β deposition, hyperphosphorylated tau accumulation, microglial activation, and neuronal loss throughout the retina, supporting direct retinal involvement in AD pathology.^11–13 Nevertheless, chronic retinal hypoperfusion resulting from microvascular rarefaction may also contribute substantially to degeneration of metabolically active retinal neurons. Dysfunction of the retinal neurovascular unit—including endothelial cells, pericytes, Müller glia, astrocytes, and neurons—may impair autoregulation, reduce oxygen and nutrient delivery, and promote chronic ischemic injury.⁴¹ Consequently, the observed INL and ORL thinning likely reflects the combined effects of primary AD-related neurodegeneration and secondary vascular insufficiency rather than either mechanism alone.

The retinal microvascular abnormalities observed in our cohort further support an important contribution of vascular dysfunction to AD. Reduced vessel density within both the superficial and deep vascular complexes is consistent with accumulating evidence that cerebral and retinal microvascular dysfunction occurs early during AD progression.^1,17–19,22 Because the retinal and cerebral microcirculations share a common embryological origin, similar endothelial architecture, and comparable blood–retinal and blood–brain barrier properties, retinal vascular alterations may mirror pathologic processes occurring within the cerebral microvasculature.⁸ Microvascular dysfunction has emerged as a central component of AD pathophysiology, contributing to impaired cerebral perfusion, blood–brain barrier disruption, reduced vascular reactivity, neuroinflammation, and impaired clearance of toxic metabolites.² Our observation that lower retinal vessel density was associated with cortical atrophy and cognitive impairment supports the hypothesis that retinal vascular changes represent downstream manifestations of widespread neurovascular dysfunction rather than isolated ocular abnormalities.

A major strength of the present study is the integration of retinal imaging with quantitative MRI measures. We observed that reduced GCIPL thickness and lower SVC density were predominantly associated with cortical thinning involving the temporal, inferior parietal, precuneus, supramarginal, lateral occipital, and pericalcarine cortices—regions consistently recognized as preferential targets of AD neurodegeneration.^4,42 These spatially localized associations suggest that retinal abnormalities are biologically linked to established patterns of cerebral degeneration rather than reflecting generalized aging. Although the present cross-sectional design precludes causal inference, the anatomic specificity of these associations supports the concept that retinal imaging may provide a surrogate marker of ongoing cerebral neurodegeneration.

The observed associations between retinal imaging biomarkers and plasma Aβ42 and p-tau217 further strengthen the biological relevance of retinal alterations in AD. Lower plasma Aβ42 concentrations were associated with thinning across multiple retinal layers together with reduced retinal vascular density, whereas elevated plasma p-tau217 was associated primarily with reduced ORL thickness and lower SVC density. Plasma p-tau217 is among the most accurate blood-based biomarkers of AD pathology and closely correlates with amyloid and tau PET imaging across the disease continuum. These findings therefore suggest that retinal structural and vascular abnormalities parallel molecular processes associated with both amyloid accumulation and tau-mediated neurodegeneration.²² Nevertheless, retinal imaging should be considered complementary rather than equivalent to molecular biomarkers, as OCT/OCTA measures downstream structural and vascular consequences of disease rather than directly quantifying pathologic protein deposition.

Interestingly, although APOE ε4 carriers demonstrated significantly greater GCIPL thinning, lower retinal vascular density, and more extensive cortical atrophy, plasma Aβ42 and p-tau217 concentrations did not differ significantly from noncarriers. This apparent discrepancy may reflect differences in the biological information captured by fluid biomarkers versus structural imaging biomarkers. Plasma biomarkers primarily reflect ongoing molecular pathology and may plateau during symptomatic stages of AD, whereas retinal imaging and MRI quantify cumulative neurodegenerative and microvascular injury that continues to evolve throughout disease progression.^43–46 In addition, APOE ε4 has important vascular effects independent of amyloid deposition, including impaired endothelial function, disruption of blood–brain and blood–retinal barrier integrity, reduced vascular reactivity, and altered neurovascular coupling.⁴⁷ These mechanisms may increase retinal susceptibility to microvascular injury and neuronal degeneration independent of measurable differences in circulating amyloid or tau concentrations, providing a plausible explanation for the more severe retinal phenotype observed among APOE ε4 carriers.

Retinal imaging biomarkers were also significantly associated with global cognitive performance. Among all retinal parameters, GCIPL thickness demonstrated the strongest and most consistent relationship with both MMSE and MoCA scores, whereas reduced SVC density was associated primarily with MoCA performance. These findings suggest that retinal neuronal degeneration may be more closely linked to global cognitive decline than retinal vascular alterations alone. Nevertheless, our interaction analyses demonstrated that retinal microvascular integrity modified the relationship between cortical thickness and cognition, supporting the concept that neuronal degeneration and vascular dysfunction jointly contribute to cognitive deterioration in AD. Because cognition is influenced by numerous interrelated pathologic processes, retinal biomarkers should be considered components of an integrated multimodal assessment rather than independent predictors of cognitive impairment.

Molecular neuroimaging remains the reference standard for in vivo assessment of AD pathology.^2,42,48,49 Amyloid and tau PET imaging provide highly specific measures of cerebral protein deposition and play a central role within the current biological framework of AD diagnosis. However, PET imaging is expensive, not universally available, and exposes patients to ionizing radiation. In contrast, OCT/OCTA is rapid, noninvasive, radiation-free, and widely available in routine ophthalmic practice. Rather than replacing molecular imaging, retinal imaging may complement established biomarkers by providing a scalable assessment of downstream neurovascular and neurodegenerative changes. Such an approach may be particularly valuable for population screening, longitudinal monitoring, and identifying individuals who would benefit from more comprehensive biomarker evaluation using PET or cerebrospinal fluid analyses.

The relationship between the eye and brain extends beyond their anatomic continuity through the optic nerve and encompasses shared developmental origins, neuronal characteristics, vascular architecture, and neurovascular unit components. The retina may therefore provide a readily accessible window into processes occurring within the central nervous system. In the context of AD, retinal neurodegeneration and microvascular dysfunction may reflect convergent effects of amyloid- and tau-related pathology, vascular dysfunction, and impaired neurovascular coupling. Our multimodal findings, demonstrating coordinated associations among retinal abnormalities, cerebral atrophy, plasma AD biomarkers, and cognitive impairment, further support the potential of retinal imaging as a component of the broader eye–brain axis.⁵⁰ Although our findings support the translational potential of retinal imaging, several limitations should be acknowledged. First, the cross-sectional design precludes determination of temporal or causal relationships among retinal alterations, cerebral degeneration, molecular pathology, and cognitive decline. Longitudinal studies are required to establish whether retinal biomarkers precede clinical progression or track therapeutic response. Second, although major vascular risk factors were adjusted for statistically, residual confounding from systemic vascular health cannot be completely excluded. Third, we evaluated global retinal thickness and vessel density rather than regional sectoral measurements, which may detect localized retinal changes with greater sensitivity. Fourth, retinal vascular measurements may vary across OCTA platforms because of differences in segmentation algorithms, particularly regarding the definition of superficial and deep vascular complexes. Although all participants were imaged using the same validated swept-source OCTA system, direct quantitative comparison with studies employing alternative devices should therefore be interpreted cautiously. Fifth, although rigorous image-quality control was performed according to OSCAR-MP and APOSTEL recommendations, segmentation accuracy may be reduced in eyes with advanced retinal thinning, potentially introducing minor measurement variability. Finally, this was a single-center study comprising exclusively Chinese participants imaged on a single OCT/OCTA platform. Because retinal pigmentation, ocular biometry, and retinal vascular architecture vary across ethnic populations and imaging systems, external validation in larger, ethnically diverse, multicenter cohorts is necessary before widespread clinical implementation.

Conclusions

Retinal structural and microvascular abnormalities measured by OCT/OCTA were closely associated with cortical atrophy, plasma biomarkers of AD pathology, APOE ε4 status, and cognitive impairment. These findings support the retina as an accessible window into cerebral neurodegeneration and microvascular dysfunction. While retinal imaging is unlikely to replace established molecular biomarkers, it offers considerable potential as a rapid, noninvasive, and scalable adjunct for risk stratification, disease monitoring, and multimodal assessment across the AD continuum.