Hyperspectral retinal imaging as a screening tool for Alzheimer’s disease

Abstract

Purpose : The recent approval by the United States Food and Drug Administration for anti-amyloid therapy in early-stage Alzheimer's disease (AD) has highlighted the importance of an early diagnosis in treating the disease before symptoms manifest. Optina Diagnostics' Mydriatic Hyperspectral Retinal Camera (MHRC) has the potential to diagnose AD by identifying cerebral amyloid-associated image features in the retina through an easy and non-invasive fundus scan. This study aims to identify spatial-spectral retinal features correlating with the presence of Amyloid beta (Aβ) in the brain.

Methods : The MHRC obtained retinal images from 10 healthy controls with negative amyloid positron emission tomography (PET) scans and 32 participants with positive amyloid PET scans. Hyperspectral data cubes, each comprising 92 retinal images with 5 nm increments across the 450-905 nm spectral range and a 31° field of view, were generated in approximately one second, providing information on the spatial and spectral reflectance of retinal tissue. Spatial-spectral features were extracted from three cubes per participant using various combinations of anatomical masks, spectral regions, and texture measures. These features were assessed for statistical significance using Tukey's test to classify the cerebral amyloid status determined by amyloid PET scans. Features were considered significant if P < 0.05 for both our current cohort (n=42) and an independent training cohort (n=522).

Results : We extracted 2304 spatial-spectral features, with 17 identified as significant (P value range of 0.0114-0.0465) for classifying amyloid PET status. An additional 935 morphological features relating to the diameter, tortuosity, density, and fractal dimension of blood vessels from different retinal zones were extracted. Five morphological features were also significant (P value range of 0.0173-0.0395) for classifying amyloid PET positivity.

Conclusions : Spatial-spectral features extracted from hyperspectral retinal images show promise in identifying individuals with AD, surpassing the morphological features available through conventional retinal imaging. An algorithm is currently under development to classify features using machine learning to discriminate between Aβ-positive and Aβ-negative individuals. The MHRC has the potential to offer a non-invasive, patient-friendly, and cost-effective screening tool for AD.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.