Optical coherence tomography angiography for assessing retinal microvascular changes in Alzheimer's disease: the potential of retinal metrics in the asymptomatic stage

Abstract

Alzheimer’s disease (AD) presents a significant global health challenge. AD is usually diagnosed in its advanced stages after permanent cerebral damage, making it very difficult to treat. Microvascular dysfunction is a crucial aspect of dementia and is believed to be a driving factor in its progression. If quantifiable metrics for the early stage could be identified, it would contribute to a more accurate estimation of an individual’s risk of developing the disease. While existing biomarkers such as cerebrospinal fluid (CSF) Aβ42 amyloid, medial temporal-lobe atrophy, and white matter lesion volumes aid in identifying and monitoring symptomatic AD, they have not been able to reliably identify AD in its early stage.

The retina is linked to the brain’s cognitive areas and shares embryological origin. Optical coherence tomography angiography (OCTA) is a non-invasive imaging modality that visualizes the retinal microvasculature in high resolution. Being a non-invasive imaging modality has consolidated OCTA as a valuable tool for detecting vascular changes related to the retina and brain diseases such as cerebral small vessel disease and symptomatic AD. This sets a solid foundation for the extension of OCTA research to asymptomatic AD.

This thesis used a previously published framework designed to extract quantitative metrics from OCTA images that characterize the retinal microvasculature through various retinal metrics. To further validate the framework for use in this thesis, it was important to establish the reproducibility of the framework. The first analysis investigated whether previously reported retinal metrics computed from the framework were reproducible across repeated scans of 22 eyes from 22 unique patients (mean age = 64.86+/-7.30 years, 77% female). Furthermore, the impact of different vessel segmentation methods was investigated on the reproducibility of these metrics (optimally oriented flux (OOF) and Frangi filter segmentation). Results demonstrated the reproducibility of OCTA retinal metrics across repeated scans and suggest that the choice of vessel segmentation method can significantly impact the reproducibility of OCTA retinal metrics.

Once reproducibility of the framework was established, retinal microvascular changes in midlife participants at risk of developing AD later in life were investigated. The aim was to identify OCTA retinal metrics that show a significant difference between control and participants at risk of developing AD (mean age = 51.2 years). Three different risk groups were investigated from a single cohort study. Participants with and without the Apolipoprotein E4 (APOE4) gene, participants with and without a known family history of dementia (FH), and participants with high and low Cardiovascular Risk Factors, Aging, and Dementia (CAIDE) scores. This was carried out at baseline and over a two-year follow-up. Results demonstrated that there were retinal metrics that showed differences between control and participants at risk of developing AD across risk groups at baseline and over a two-year follow-up.

In addition to investigating retinal microvascular changes in mid-life individuals at risk of developing AD, it was important to explore the relationship between retinal microvasculature changes and established neuroimaging biomarkers for symptomatic AD. Results demonstrated significant associations between OCTA retinal metrics and increased WMH burden across various brain regions. This included the whole brain, deep, and periventricular areas. These significant associations carried over across the brain’s four lobes, including the left and right occipital, frontal, temporal, and frontal lobes. The thesis also investigated whether there were potential correlations between retinal microvasculature changes and gray matter volumes (GMV). There was a significant association between OCTA retinal metrics and the whole brain, subcortical, and left and right hippocampus.

The final analysis on retinal microvascular changes investigated how OCTA retinal metrics progress between a mid-life cohort that is at risk of developing AD to an elderly cohort that consists of individuals diagnosed with mild cognitive impairment (MCI) and AD. To explore the progression of OCTA retinal metrics, the analyses leveraged the use of two distinct cohorts: the PREVENT cohort with participants at risk of developing AD and Duke University’s study cohort with participants consisting of a diagnosis of elderly controls, MCI and AD. The findings revealed that a retinal metric related to vascular density was significantly different between mid-life at-risk participants (high CAIDE score) and elderly participants with AD and MCI even when the same metric remained constant between the control groups of both cohorts. This highlights the potential of OCTA for tracking the progression of AD from asymptomatic to symptomatic stages.

The scarcity of publicly available OCTA datasets and privacy concerns with sharing real patient data hinder the development of robust diagnostic models. This thesis proposed the use of Latent Diffusion Models (LDM) for generating novel and anatomically accurate synthetic OCTA images as a method for sharing health-protected OCTA datasets. Anatomical accuracy was confirmed as there were no significant differences in the OCTA retinal metrics between the synthetic and real images in both conditions (AD and control). A Logistic Regression model trained only on synthetic retinal metrics achieved comparable AUC scores (AUC: 0.614, CI: 0.546, 0.668) to the model trained only on real metrics (AUC: 0.616, CI: 0.546, 0.642). Increasing the number of synthetic samples improved the model’s performance and robustness (AUC of 0.634 CI: 0.618, 0.654 using 3500 synthetic samples). LDMs can generate synthetic OCTA images that encapsulate diagnostic information present within real images.

In conclusion, this thesis contains novel analyses that add to growing evidence on the potential use of OCTA as a valuable tool for detecting early vascular changes associated with AD. Further research should be conducted to validate the retinal metrics as reliable biomarkers for early-stage detection.