Application of Automated Quantification of Fluid Volumes to Anti-VEGF Therapy of Neovascular Age-Related Macular Degeneration

Purpose: Anti-vascular endothelial growth factor (VEGF) treatment of neovascular age-related macular degeneration (AMD) is a highly effective advance in the retinal armentarium. OCT offering 3-dimensional imaging of the retina is widely used to guide treatment. Although poor outcomes reported from clinical practice are multifactorial, availability of reliable, reproducible, and quantitative evaluation tools to accurately measure the fluid response, that is, a VEGF meter, may be a better means of monitoring and treating than the current purely qualitative evaluation used in clinical practice. Design: Post hoc analysis of a phase III, randomized, multicenter study. Participants: Study eyes of 1095 treatment-naive subjects receiving pro re nata (PRN) or monthly ranibizumab therapy according to protocol-specified criteria in the HARBOR study. Methods: A deep learning method for localization and quantification of fluid in all retinal compartments was applied for automated segmentation of fluid with every voxel classified by a convolutional neural network (CNN). Three-dimensional volumes (nanoliters) for intraretinal fluid (IRF), subretinal fluid (SRF), and pigment epithelial detachment (PED) were determined in 24 362 volume scans obtained from 1095 patients treated over 24 months in a phase III clinical trial with randomization to 2 drug dosages (0.5 mg and 2.0 mg ranibizumab) and 2 regimens (monthly and PRN). A multivariable mixed-effects regression model was used to test for differences in fluid between the arms and for fluid/function correlation. Main Outcome Measures: Fluid volume in nanoliters, structure-function as Pearson's correlation coefficient, and as a coefficient of determination (R-2). Results: Fluid volumes were quantified in all visits of all patients. Automated segmentation demonstrated characteristic response patterns for each fluid compartment individually: Intraretinal fluid showed the greatest and most rapid resolution, followed by SRF and PED the least. The loading dose treatment achieved resolution of all fluid types close to the lowest levels attainable. Dosage and regimen parameters correlated directly with resulting fluid volumes. Fluid/function correlation showed a volume-dependent negative impact of IRF on vision and weak positive prognostic effect of SRF. Conclusions: Automated quantification of the fluid response may improve therapeutic management of neovascular AMD, avoid discrepancies between clinicians/investigators, and establish structure/function correlations. (C) 2020 by the American Academy of Ophthalmology.