Abstract
Purpose :
To investigate treatment effects on local progression of Geographic Atrophy (GA) with respect to topographic growth patterns quantified from SD-OCT images by deep learning in age-related macular degeneration (AMD).
Methods :
334 SD-OCT scans of 57 eyes with monthly (AM), 52 eyes with every-other-month (AEOM) treatment and 58 eyes with sham (SM) injection from baseline and 1 year follow up of a Phase 2 clinical trial evaluating pegcetacoplan in patients with GA secondary to AMD (FILLY, NCT02503332) were included. Retinal pigment epithelium loss were automatically segmented using deep learning. Local GA growth rate was determined from the delineated en-face projections of GA as the distance of baseline border points to the closest border point at 1 year. Growth direction was categorized into “towards fovea” when angle between growth direction and direction to fovea was less than ±90° and “away from fovea” otherwise.
Mixed effects models grouped by eyes and distance as random factor were used to regress growth speed for each GA lesion border point using treatment and growth direction as covariates.
Results :
Using 321,946 GA border points, mean growth for SM was 64 µm/year (95%CI: 39 to 87), with 18 µm/year less(95% CI: 5 to 32, p=.0103) and 6 µm/year less(95% CI: -8 to 20, p=.39) for AM and AEOM, respectively. For the 195,396 points showing GA growth, the mean growth was 108µm/year (95%CI: 86 to 132) for SM and additional reduced growth by 26 µm/year (95%CI: 6 to 47, p=.0135) and 4 µm(95%CI: -16 to 25, p=0.72) for AM and AEOM, respectively. Growth towards fovea is reduced by 4 µm (95%CI: 2 to 5, p=<.001) for SM and additionaly by 13µm (95%CI: 11 to 15, p<0.0001) and 12 µm (95%CI: 11 to 14, p<0.0001) for AM and AEOM treatment.
Conclusions :
Eyes treated with pegcetacoplan showed a significantly slower GA lesion growth rate compared to sham, and an even slower growth rate towards fovea. By assessing local GA growth and direction, the heterogeneity in GA progression is captured in more detail and enables observation of treatment effects with respect to topography. Using deep learning to automatically quantify atrophy in in-vivo OCT images allows for precise and accurate assessment and analysis of GA progression at a large scale and may provide new insight into disease mechanisms and response to novel treatment.
This is a 2021 ARVO Annual Meeting abstract.