Retinal Microvascular Alterations in Alzheimer’s Disease and Mild Cognitive Impairment

Background: The retina and brain share many neuronal and vasculature characteristics developmentally and potential biomarkers may be present in the retina. We investigated the retinal microvasculature in Alzheimer’s disease (AD) and mild cognitive impairment (MCI) using optical coherence tomography angiography (OCTA). Methods: In this cross-sectional study, 24 AD participants, 37 MCI participants, and 29 controls were diagnosed according to internationally accepted criteria. OCTA images of the supercial and deep capillary plexus (SCP, DCP) of the retinal microvasculature were obtained using a commercial OCTA system (Zeiss Cirrus HD-5000 with AngioPlex, Carl Zeiss Meditec, Dublin, CA). The main outcome measures were vessel density (VD) and fractal dimension (FD) in the SCP and DCP within a 2.5-mm ring around the fovea were compared between groups. Perfusion density of large vessels and foveal avascular zone (FAZ) area were additional outcome parameters. Results: Age, gender and race did not differ among groups. However, there was a signicant difference in diabetes status (P=0.039), and systolic blood pressure (P=0.008) among the groups. After adjusting for confounders, AD participants showed signicantly decreased VD in SCP and DCP (P = 0.005 and P = 0.016, respectively) and decreased FD in SCP (P = 0.008), compared to controls. MCI participants showed signicantly decreased VD and FD only in SCP (P = 0.005 and P < 0.001, respectively) and not the DCP (P > 0.05) compared with controls. There was no difference in the OCTA variables between AD and MCI (P > 0.05). Perfusion density of large vessels and FAZ area did not differ signicantly between groups (P > 0.05). Conclusions and relevance: Eyes of patients with AD have signicantly reduced macular VD in both plexuses whereas MCI participants only showed reduction in the supercial plexus. Changes in the retinal microvasculature and capillary network may mirror small vessel cerebrovascular changes in AD.


Introduction
Alzheimer disease (AD) is a signi cant cause of dementia and has important implications for patients and their families. Globally, the number of individuals living with dementia is set to rise, particularly in low-and middle-income countries. [1] The retina and brain shares many neuronal and vasculature characteristics [2] and potential biomarkers may be present in the retina. Previous studies have analyzed digital fundus photographs and reported a range of retinal vessel alterations in patients with AD and mild cognitive impairment (MCI). [3] However, images obtained from this technique can only provide information of retinal arterioles and venules measuring 60-300 µm in diameter. [4] Optical coherence tomography angiography (OCTA) is a recent innovation that allows for further quanti cation of the retinal microvasculature and visualization of capillaries measuring 5-15 µm in diameter, which may be more representative of the entire microvascular network. [5,6] Thus, the OCTA may be a potential non-invasive optical imaging tool to determine the presence and role of microvascular dysfunction in AD and cognitive impairment.
Furthermore, OCTA is capable of measuring retinal capillary beds at distinct depths, separating the super cial capillary plexuses (SCP) and deep capillary plexuses (DCP), each re ecting the metabolic demand of particular neuronal layers. [5] In AD, the tissue of interest is the inner retinal layer, as re ected by the loss of retinal ganglion cells, [7,8] thinning of the retinal nerve bre layer thickness and ganglion cell layer thickness, [9] and deposition of β-amyloid (Aβ) plaques. [10] While there are a few OCTA studies investigating AD, there have been mixed conclusions. [11][12][13][14][15] Some researchers reported nding signi cant reduction in the vessel density (VD) only in the super cial plexus, [14,16] which complements histology ndings [10] and OCT studies [9] since the super cial plexus mainly supplies the inner retinal layer. [17] However, others reported nding changes only in the deep plexus. [11,15] Studies [15,16] have also used OCTA to examine participants with MCI, who are at higher risk for dementia and AD, but have drawn con icting results as well. For example, while Zhang et al. [16] found signi cantly decreased VD only in the super cial plexus, Wu et al. [15] found a reduction in the VD only in the deep plexus. Therefore, it remains unclear whether there is a difference in the OCTA parameters in AD and MCI individuals, partly due to the physiologic variability of the foveal avascular zone (FAZ), [18,19] and projection artefacts of the DCP. [20,21] To address these gaps, the purpose of the current study is to compare the retinal microvasculature metrics using OCTA, accounting for potential measurement bias of FAZ and projection artefacts of DCP in participants with AD, MCI, and controls. We hypothesize that alterations in OCTA metrics as characterized by sparser vessel density and loss of vessel complexity in the retinal vasculature, occur in AD and to a lesser extent in MCI compared to controls.

Study participants
We conducted a cross-sectional case-control study, approved by National Healthcare Group Pte Ltd Domain Speci c Review Board and adhered to the Declaration of Helsinki. Written informed consent was obtained from each participant or their primary caregiver before enrollment.
AD and MCI individuals: Participants aged 50 years of age or older were enrolled from an ongoing longitudinal memory clinic-based study from December 2018 to October 2019. Weekly consensus meetings were held with study clinicians, neuropsychologists, clinical research fellows, research coordinators and research assistants. Details from the clinical assessment, blood investigations, neuropsychological testing and MRI scans were reviewed. AD participants were diagnosed in accordance with the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV criteria. [22] MCI participants were those who were de ned as impairment in 1 or more domains in the neuropsychological test battery that does not affect activities of daily living. Controls: Individuals who attended the same clinics but were not impaired in any of the cognitive domains tested. Medical histories (e.g. diabetes and hypertension) were collected and seated blood pressure (BP) measurements were taken using an automated oscillometric device during their clinical visits.

Ocular examinations
After pupil dilation with 1% tropicamide and 2.5% phenylephrine hydrochloride, each participant underwent ocular imaging in both eyes that included retinal photography with a nonmydriatic digital camera and OCTA imaging. Two retinal fundus photographs, with one centered at the optic disc and another centered at the macula, were obtained to document the absence of eye diseases.
Optical coherence tomography angiography All OCTA scans were performed by a single trained technician, using the Zeiss Cirrus HD-5000 Spectral-Domain OCT with AngioPlex OCTA (Carl Zeiss Meditec, Dublin, CA), that featured a central wavelength of 840 nm, a speed of 68,000 A-scan per second and an axial resolution of 5 µm and transverse resolution of 15 µm in tissue. The FastTrac motion correction, based on a line-scanning ophthalmoscope, was enabled to minimize motion artefacts during acquisition. Each participant received a 3 × 3-mm 2 scan, with each scan consisting of an isotropic sampling (245 × 245) and four consecutive B-scans obtained at each raster location to compute the angiographic information using an optical microangiography protocol. [23] A trained grader masked to the participant's characteristics reviewed the quality of all OCTA scans. All Bscans were checked for alignment and segmentation errors. We excluded participants from the analysis if the OCTA images from both eyes were of poor quality (poor signal strength index < 7, signi cant motion artefacts visible as irregular vessel patterns on the en face angiogram, local weak signal caused by artefacts such as oaters, misalignment or incorrect segmentation). [24,25] A randomly selected eye was analyzed for each participant since measurements of both eyes were highly correlated.
Each scan was automatically segmented into the super cial capillary plexus (SCP) and deep capillary plexus (DCP) by a review software (Carl Zeiss Meditec, version 11.0.0.29946). The SCP spans the inner limiting membrane (ILM) to the inner plexiform layer (IPL), while the DCP spans the inner nuclear layer (INL) to the outer plexiform layer (OPL). [26] Images were checked to ensure correct segmentation by the automated instrument software and no manual adjustment was needed. Projection artefacts from the overlying retinal circulation were removed from the DCP using the removal software that was integrated with the instrument.
OCTA images of the super cial and deep retinal plexuses were loaded into a customized algorithm using MATLAB (The MathWorks Inc., Natick, MA). The framework of OCTA image processing involved the following steps ( Fig. 1): (1) manually outlined the border of the foveal avascular zone (FAZ) of the super cial and deep vascular plexus angiograms; [27] (2) applied a Hessian-based lter to enhance the contrast of large vessels on the SCP, which is consequently (3) binarized; [28] , [29] (4) performed the region-based analysis with a fovea-centered annulus that has an inner diameter of 1 mm and outer diameter of 2.5 mm. (5) VD was de ned as the total length in mm of perfused retinal microvasculature per unit area in in the annulus region of measurement. Perfusion density of the large vessels were computed as the ratio between large vessel area per total imaged area in the annulus region of measurement. (6) FD represents the vessel complexity of the retinal vasculature [30] and was calculated within the annulus zone using the box counting method (Dbox) with the fractal analysis toolbox (TruSoft

Results
Of the 166 participants who were enrolled and imaged between December 2018 to October 2019, we excluded participants who were unable to complete OCTA scanning due to fatigue (n = 15), poor scan quality (n = 54), and presence of eye diseases such as glaucoma, vascular or nonvascular retinopathies, age-related macular degeneration since this could potentially confound the OCTA results (n = 7), leaving 24 AD participants, 37 MCI participants, and 29 control participants with good quality OCTA for analysis.
There was no signi cant difference in age, gender and race, among the groups ( Table 1). The mean ± standard deviation (SD) age of participants was 76.8 ± 6.0 years, 51% were female and 81% were Chinese. Of note, AD participants had higher diabetes prevalence (P = 0.039) and higher levels of systolic BP (P = 0.008) than participants with MCI or controls. There was a small but signi cant difference in the average OCTA quality in participants with AD (9.8 ± 0.8 signal strength), MCI participants (9.8 ± 0.5 signal strength) and controls (9.6 ± 0.7 signal strength; P = 0.010).  Alzheimer's disease (AD) and mild cognitive impairment (MCI) Data presented are mean (SD) or number (%), as appropriate.
* P value was obtained with ANOVA for the continuous variables and with chi-square tests for categorical variables.
After adjusting for age, gender, race, diabetes and blood pressure (systolic and diastolic levels), AD participants showed signi cantly decreased VD in SCP and DCP (P = 0.005 and P = 0.016, respectively) and decreased FD in SCP (P = 0.008; Table 2), compared to controls. MCI participants showed signi cantly decreased VD and FD only in SCP (P = 0.005 and P < 0.001, respectively) and not the DCP (P > 0.05) compared with controls. Figure 2 further illustrates the VD and FD among the groups. There were no statistically signi cant differences in the OCTA variables between AD and MCI (P > 0.05). Perfusion density of large vessels and FAZ area did not differ signi cantly between groups (P > 0.05; Table 3).

Discussion
In this cross-sectional study, we evaluated the extent and pattern of retinal microvascular alterations, speci cally at the capillary network level, in AD and MCI. We compared three retinal OCTA metrics (VD, FD and FAZ area) in two capillary plexuses (super cial and deep) in AD, MCI, and controls. Compared with controls, AD participants showed signi cantly sparser VD in both plexuses whereas MCI participants only showed reduction at the super cial plexus. In terms of FD, AD and MCI participants exhibited a loss of vessel complexity of the SCP when compared with controls. Our study adds further to the concept that there are possible progressive differences in retinal microvascular alterations between AD and MCI; the use of VD in the SCP (together with DCP) may further distinguish between AD (both SCP and DCP affected) and MCI (only SCP is affected) individuals. Taken together with increasing evidence from other research, our current study demonstrates that differences in retinal microvascular changes using OCTA may potentially be used to identify and screen for AD and earlier cognitive phenotypes (i.e., MCI).

SCP in AD individuals
We showed that AD participants have a sparser VD of the SCP compared with control participants. Our ndings support most of the previous OCTA studies in AD participants (Table 4), [12][13][14]16] which is in keeping with studies on larger retinal vessels using fundus photographs. [3,33] It should be noted that two other studies did not observe any differences in the VD of the SCP in AD participants. [11,15] OCTA quanti cation metrics may potentially be affected by several confounders. First, OCTA signal strength quality can affect the VD, where the VD decreased linearly with signal strength. [34] In our study, although there was a small difference in the OCTA signal strength quality between controls and AD/MCI (9.6 vs 9.8 out of 10), the scan qualities were extremely high. Second, although studies have excluded participants with uncontrolled hypertension, the BP levels can affect the VD. [25] In the current study, we statistically adjusted the BP levels to remove the BP bias. Last, the physiological variability of FAZ can affect the VD. [35] This is mainly dependent on how much FAZ one includes in the analytical regions. In eyes with a larger FAZ, the FAZ would occupy a larger portion of the analytical area, resulting in a lower VD. In the current study, we mitigated the potential measurement bias by manually delineating the FAZ region and masking it from the calculation.  The SCP is responsible for the metabolic supply of the ganglion cell layer, [17] where reduced number of retinal ganglion cells and axons has also been observed in post-mortem AD retinas. [7,8] Changes in the SCP seen on OCTA further complement the already established retinal OCT structural markers. [9] Using OCT, several studies have reported the reduction of retinal nerve bre layer thickness and ganglion cell layer thickness (presumably due to loss of retinal ganglion cells and axonal degeneration) in AD patients. [9] Whether the loss of retinal vessel precedes the loss of retinal neurons is currently unknown.

SCP in MCI individuals
Discordant results have been reported on the OCTA ndings in MCI participants ( Table 3). The current and one previous study [16] showed a reduced VD of the SCP in those with MCI whereas other studies did not report any differences. [13,15] Alteration in the retinal vessels in MCI participants is compatible with studies using in vivo Doppler imaging techniques, where a decrease in retinal blood ow has been demonstrated in both AD and MCI participants. [36,37] The con icting results between MCI and control participants may lie with the de nition of MCI, which represents a continuum of cognitive decline between "normal aging" and dementia. While the person is still able to carry out their activities of daily living with little or no help from others, a wide range of cognitive impairment is possible in MCI. [38] It is plausible that the change in retinal capillaries may occur only at a more severe stage of MCI or when certain cognitive ability is affected.

DCP in AD individuals
Four OCTA studies have investigated DCP VD in AD individuals, but there is generally a lack of agreement between studies (Table 4). Two studies [11,15] showed a signi cant reduction in VD in AD individuals whereas the others [14,16] did not observe any differences. Obtaining accurate OCTA metrics from the DCP layer is particularly challenging as it is affected by the physiologic variability of FAZ [18,19] and projection artefacts. [20,21] First, previous OCTA studies did not account for the FAZ in the deeper plexus.
This is crucial because the FAZ in the deep plexus is considerably larger than super cial plexus. [39] Second, while Zabel et al. removed the projection artefacts in the DCP, [11] the rest did not. [14][15][16] In the current study, we quanti ed the VD of the DCP without the in uence of FAZ and projection artefacts, which hopefully reduced measurement bias. We found a sparser VD of the DCP in AD participants but not in MCI individuals, which suggests the possibility of using the VD of the DCP to discriminate between (both SCP and DCP affected) and MCI (only SCP is affected) individuals.
The DCP is important for nutrition of the inner nuclear layer, which comprises of bipolar cells, horizontal cells, and amacrine cells. [17] In transgenic AD mouse models, Aβ deposits have been detected in the inner nuclear layer. [10] Microvascular changes of the DCP are complemented by the structural thinning of the inner nuclear layer thickness in AD individuals. [40] Changes in the DCP in AD individuals may present later in the disease stage, but a longitudinal study will be required to con rm this hypothesis.

Fractal dimension
In addition to capillary loss, we saw a signi cantly decreased FD of the SCP, which suggests a loss of vessel complexity in the inner retinal macula of those with AD and MCI compared to cognitively normal controls. However, an earlier study [41] observed a signi cant reduction in the FD in both plexuses in AD and MCI individuals. In contrast, we did not nd any difference in the FD of the DCP between the groups.
A plausible reason for the disagreement may be related to the vascular arborization pattern of the distinct layers. The SCP is supplied by the central retinal artery and composed of vessels running parallel of the retinal surface, thereby displaying a distinct vascular tree whereas the DCP is supplied by vertical anastomoses from the SCP, presenting as a lobular con guration. [17] Since FD is a measure of vasculature branching pattern complexity, it may be a more relevant biomarker for the SCP than the DCP.
Our nding is in keeping with previous publications on larger retinal vessels using fundus photographs, where AD and MCI participants demonstrated a loss of vessel complexity. [30] Perfusion density of large vessels It should be noted that although VD in SCP was decreased in participants with cognitive impairment, perfusion density of the large retinal vessels remained unchanged. Previous studies have examined retinal vessels from fundus photographs and reported alterations in the venular caliber in participants with AD. [3] Retinal vessels measured from fundus photos are considerably larger in diameter than those obtained from OCTA. [4] Also, the diameters of retinal vessels measured with OCTA are in good agreement with the ground truth as obtained with adaptive optics ophthalmoscope. [42] Therefore, the lack of large vessel changes despite capillary changes suggests that microvascular alterations precede large vessel changes. The capillaries in the SCP may be particularly susceptible to the deleterious effects of neurodegeneration whereas the large retinal vessels may change later in the pathogenesis of AD. This nding would also suggest that the use of OCTA may be more sensitive in detecting changes in AD and MCI participants than fundus camera.

Foveal avascular zone area
Previous OCTA studies have quanti ed the FAZ area within the SCP region automatically using the OCTA software and reported con icting results (Table 4). [11][12][13][14][15] Some reported a signi cant enlargement of the FAZ in individuals with AD compared to controls, [11,12,15] whereas others reported no difference in the FAZ area. [13,14] In the current study, we did not nd any differences in either of the plexuses between groups. Overall, the FAZ area has numerous limitations to serve as a biomarker of cognitive impairment given its physiologic variability, effect of axial length on OCT scan dimensions and segmentation/measurement limitations. [27,35,[43][44][45] Strengths and limitations Strengths of this study include a well-phenotype cohort of AD and MCI individuals who were diagnosed according to internationally accepted criteria, and a standardized study methodology which further improved the validity of the imaging data. As explained above, we accounted for larger retinal vessels, FAZ dimension and projection artefacts from the analysis. Second, a quarter of our participants were excluded because of poor quality OCTA scans. Such high exclusion is comparable to other OCTA studies (~ 22% OCTA scans were rejected). [13] Since OCTA is based on motion detection, it is particularly sensitive to the patient's eye movement. The need for good patient xation can be challenging in elderly patients with cognitive impairment. Incorporating eye tracking during OCTA scanning can lessen the eye movement [46] but may lead to longer image acquisition time, which in turn result in patient fatigue. Nonetheless, we performed rigorous quality control on all scans, where all B-scans were checked for misalignment and segmentation errors. Images with segmentation errors, poor signal strength or eye diseases, which might potentially affect the OCTA measurements, were excluded. This robust data preparation is also evident by the particularly high signal strength among our participants, and further corroborates the validity of our results. Third, we adjusted the possible confounding effects of age, gender, race, and the presence of diabetes, and systemic blood pressure levels during multivariate analysis and excluded possible confounding factors, such as glaucoma, vascular or nonvascular retinopathies, age-related macular degeneration. [24,25,32] Our present study has a few limitations. First, even though there was reduced retinal VD in those with AD and MCI using OCTA, this was only a cross-sectional study. It remains unclear whether the changes of retinal capillary are predictive of cognitive decline. This will be investigated in ongoing follow-up studies.
Second, quanti cation of VD [18] and FAZ [45] can be affected by the effect of OCT magni cation. We did not perform ocular biometry and thus were not able to rescale the scan dimensions. [18] Instead, we masked the FAZ from the VD metrics which should have mitigated some of this measurement bias for the VD maps. We acknowledged that the FAZ area is still affected by the individual differences in axial length. Another limitation of this study is the relatively small sample size of those with AD and MCI, limiting the power for evaluating the OCTA metrics in different types of MCI and AD. Lastly, this study was restricted to older adults of Asian ethnicities, therefore generalizability of our results to persons with young-onset dementia or of non-Asian ethnicities may be limited. Also, it remains unclear how the OCTA relates with magnetic resonance imaging (MRI) markers of cerebrovascular disease markers and amyloid PET and CSF biomarkers.
In conclusion, our study shows that compared to cognitively normal controls, eyes of participants with AD demonstrated signi cantly sparser VD in both plexuses whereas MCI participants only showed reduction of the SCP. Furthermore, we found that AD and MCI participants exhibited a loss of vessel complexity of the SCP when compared with controls. Our ndings suggest that the changes at the retinal microvasculature at the capillary level may re ect similar changes in the cerebral vasculature in individuals with neurodegenerative diseases and demonstrates the potential of OCTA for the early screening of cognitively impaired individuals.