Utilizing Advanced Structural and Functional Features to Predict Future 5-Year Visual Acuity in Eyes with Non-neovascular Age-related Macular Degeneration Patients

Abstract

Purpose : Advanced non-neovascular age-related macular degeneration (dry AMD) is a leading cause of vision acuity (VA) loss. There are currently limited approaches to predicting future VA. The ability to link structural imaging biomarkers to function could be particularly useful in therapeutic development and clinical trial endpoints. As a result, there is an interest in investigating the connection between OCT structural features and VA in the context of dry AMD. The purpose of this analysis was to assess the feasibility of using a machine learning (ML) model to predict future visual acuity based on a combination of baseline visual acuity and an OCT-based advanced quantitative feature assessment in dry AMD patients.

Methods : This was an IRB-approved retrospective longitudinal image analysis study evaluating eyes with dry AMD. A ML-enabled multi-layer segmentation platform was utilized to identify multiple layers, including the internal limiting membrane (ILM), Bruch's membrane (BM), ellipsoid zone (EZ), and retinal pigment epithelium (RPE) and validated with expert trained readers. This allowed for extraction of multiple parameters related to photoreceptor integrity and subRPE compartment features (e.g., drusen, atrophy). Specific parameters of interest included assessment of macular burden of drusen 50 microns or greater, partial and total macular EZ attenuation burden, and area of geographic atrophy. A Support Vector Regression model trained on the initial dataset and tested on a self-generated validation set.

Results : For this analysis, 112 eyes for 83 patients with 5 year follow-up were included. Following model development, the feasibility of 5 years BCVA prediction on OCT metrics was demonstrated. The Root Mean Squared Error value associated with this model which is tested on validation set was 5.8 letters. Permutation importance feature assessment demonstrated that baseline visual acuity, EZ integrity and drusen burden at baseline where all important contributors to prediction.

Conclusions : This preliminary assessment provides evidence that advanced quantitative structural features on OCT may provide prognostic importance for predicting future VA in dry AMD. Future work will focus on performance optimization and validation

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.