Association of imaging factors derived from convolutional neural network with visual outcomes in age-related macular degeneration and polypoidal choroidal vasculopathy

We investigated the association of visual outcome in typical neovascular age-related macular degeneration (nAMD) and polypoidal choroidal vasculopathy (PCV) with or without pachychoroid with lesion areas on optical coherence tomography (OCT) quantified by convolutional neural network (CNN) analysis. Treatment-naïve 132 nAMD and 45 PCV eyes treated with ranibizumab or aflibercept for at least 12 months were retrospectively reviewed. Significant factors, including intraretinal fluid (IRF), subretinal fluid (SRF), pigment epithelial detachment (PED) and subretinal hyperreflective material (SHRM) area quantified by CNN at baseline and 12 months, were analyzed by logistic regression analyses for 3-line visual gain or maintenance of 20/30 Snellen vision. Visual gain at the final visit in nAMD was associated with a smaller SHRM at baseline (OR 0.167, P = 0.03), greater decrease in SRF and SHRM from baseline to month 12 (OR 1.564, P = 0.02; OR 12.877, P = 0.01, respectively). Visual gain in nAMD without pachychoroid was associated with a greater decrease in SRF and SHRM (OR 1.574, P = 0.03, OR 1.775, P = 0.04). No association was found in nAMD with pachychoroid and any type of PCV. Greater decrease in SRF and SHRM from baseline to month 12 was associated with favorable visual outcomes in nAMD without pachychoroid but not in nAMD with pachychoroid and PCV.