Sohn, Elliott; Zacharias, Zeb; Lee, Kyungmoo; Bach, Bernardo; Chava, Srinivas; Wang, Kai; Wu, Amy; Mukhopadhyay, Chirantan; Coussa, Razek; Han, Ian; Scheetz, Todd; Fingert, John; Abramoff, Michael; Stone, Edwin; Houtman, Jon; Sonka, Milan
1University of Iowa
Purpose: Neovascular age-related macular degeneration (nvAMD) has a complex pathogenesis, with genetics (e.g. CFH, ARMS2, and MMP9) and inflammation playing key roles. Artificial intelligence (AI) has enabled unprecedented analyses of fluid in optical coherence tomography (OCT) imaging. We determined the association of systemic inflammatory markers with visual acuity and fluid parameters derived from AI based quantitative OCT analysis of subjects with nvAMD.
Methods: An ensemble approach of U-Net architectures (convolutional neural networks) developed for retinal pathology segmentation, was trained to simultaneously segment intraretinal (IRF), subretinal fluid (SRF), and pigment epithelial detachment (PED) regions from Heidelberg OCT scans in nvAMD subjects (n=100) receiving consecutive anti-VEGF injections (comprising 6320 OCTs). 10-fold cross-validation was performed with training OCT scans from 90 subjects and a validation dataset of 10 subjects. An optimized spectral flow cytometry panels using Cytek Aurora assessed the steady-state phenotype (31-colors) and effector functions (23-colors) of human T cells from PBMCs in a subset of nvAMD subjects previously genotyped for MMP9. Using our 31-color steady-state panel, we determined the subset distributions of the main T cell populations (e.g., naïve, memory, etc.) in these subjects.
Results: 10-fold cross-validation yielded high R2 of 0.85, 0.95, and 0.89 for predicted vs ground truth of OCT quantified IRF, SRF, and PED volumes, respectively.
Flow cytometry showed that patients with high-risk MMP9 genotype have CD4 T cells with a trend (p=0.051) toward greater capacity to generate IL-4 and IL-13, both of which are major contributors to fibrosis. In nvAMD subjects receiving intravitreal anti-VEGF injections, a significant correlation of worse VA at initial, 2 years, and final visit was observed with several subsets of T cells. %CD4 Tem early-like cells showed a positive association with PED volume at subjects’ initial visit (p=0.0066).
Conclusions: A validated AI-based image analysis algorithm in nvAMD reveals association of worse fluid with higher systemic levels of inflammation. Subjects with worse visual acuity in nvAMD may experience increased levels of inflammation. Patients with high risk MMP9 genotype may be skewed to a Th2 phenotype that could be associated with dysregulated wound healing in nvAMD.
Last Updated: