Association of Mesopic Visual Sensitivities with automatically segmented Retinal Pigment Epithelium and Outer Retinal Atrophy (RORA) on Ocular Coherence Tomography (OCT)

Abstract

Purpose : Monitoring vision in age-related macular degeneration (AMD) is challenging and emerging treatments are driving demand for new functional biomarkers. To address this, we explored the association between mesopic visual sensitivities and RORA OCT-defined features in intermediate AMD, with an externally validated deep-learning segmentation model.

Methods : We analyzed 285 individuals from the Laser Intervention in Early Stages of AMD (LEAD) trial, aged 50 and above, with intermediate AMD at baseline. OCT and microperimetry were part of the study protocol. RORA features (photoreceptor degeneration [PRD], retinal pigment epithelium [RPE] loss, and choroidal hypertransmission [HT]) were segmented by the U-Net-based model from Zhang et al. on a B-scan level. PRD in isolation excluded overlapping RPE loss or HT. Segmentations were projected on the near-infrared (NIR) and matched the five sectors in the central 3.6-mm region sampled on microperimetry (figure 1). Multivariable linear mixed models explored the association of sectoral-mesopic visual sensitivity with sectoral-OCT RORA features in one model, and its individual features in another. These models were adjusted for age, OCT-sectoral drusen volume, and total reticular pseudodrusen area, calculated as previously reported in this cohort (Kumar et al, IOVS, 2022).

Results : 285 participants, 79% women and a mean age of 70±8 years, were included. Both RORA (-2.92 μm, 95%CI -3.90,-1.94, p<0.001) and its features: (PRD [-0.54 μm, 95% CI -0.85, -0.23, p=0.001] and RPE loss [-1.45μm, 95%CI -2.19, -0.70, p=<0.001]) were independently associated with sectoral visual sensitivity. HT and PRD in isolation showed no significant association (P > 0.01).

Conclusions : The increasing spatial extent of sectoral RORA on OCT as segmented by a machine learning model is associated with worsening mesopic visual sensitivities. Our findings indicate that the automatically segmented morphological features of intermediate AMD are related to function and lay the groundwork for future studies.

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

 

FIGURE 1
A) Non-segmented OCT B-scan with RORA
B) Segmented B-scan: PRD (yellow), RPE loss (green), and HT (blue). Composite of RORA and PRD in isolation in bounding boxes
C) En face projection of B-scan segmentation on NIR, matching microperimetry grid.
D) Microperimetry testing stimulus pattern combined with image 1B