Quantitative analysis of change in retinal tissues in neovascular age-related macular degeneration using artificial intelligence

Abstract

Purpose : The treatment of neovascular age-related macular degeneration (nAMD) preserves vision by reducing exudation within the retina, thus rehabilitating the retinal anatomy. We describe the change in the retinal tissues during treatment by segmenting OCT volume scans using an artificial intelligence (AI) system previously published by De Fauw et al. (2018).

Methods : Data was extracted from the Moorfields Eye Hospital AMD database for all patients that started treatment for nAMD between June 2012 and June 2017. This included segmentation data from Topcon 3D OCT-2000 scans, derived using the AI system. Only first treated eyes were used for analysis. The mean volumes and thicknesses were calculated by conversion from number of segmented voxels of each tissue class, including neurosensory retina (NR), subretinal fluid (SRF), intraretinal fluid (IRF), subretinal hyperreflective material (SHRM), hyperreflective foci (HRF), and pigment epithelial detachment (PED). The relative change in volume and thickness from baseline (month 0) to specified time points of 3, 12, and 24 months was calculated as a percentage.

Results : A total of 1982 treatment naive eyes were analysed. From baseline, the NR thickness reduced by 32.3μm (12%) at 3 months and 39.5μm (15%) by 24 months. Volume of SRF reduced by 78%, 84%, and 88% from baseline at 3, 12, and 24 months respectively. By 3 months from baseline, IRF initially reduced by 80%, however increased over time with a total reduction from baseline of 76% and 63% at 12 and 24 months. SHRM reduced by 69% in volume by 3 months, remaining relatively constant by 12 months (71% reduction) and 24 months (69% reduction). The volume of HRF reduced by 19% at 3 months, and by 69% at 12 months, remaining constant by 24 months (70% reduction). The PED volume reduced by 34% from baseline by 3 months, and by 46% at 24 months.

Conclusions : Tissues reflecting exudation such as IRF and SRF reduced in volume, with the greatest reduction observed within 3 months from commencing treatment. SRF continued to resolve at 2 years whereas IRF persisted. The reduction in volume of SHRM and HRF plateaued by 12 months. The quantitative analysis of change in different retinal tissues over time, enabled by AI segmentation, may provide valuable insights on disease progression, identify new subgroups of AMD, and advise on the most effective treatment protocol.

This is a 2020 ARVO Annual Meeting abstract.