Project Summary Diabetic retinopathy (DR) affects over 100 million people worldwide and is a leading cause of blindness in working-age adults. Currently, management relies on treatment decisions made by specialists relying on multiple imaging modalities, including examination of fundus photography images (frequently difficult to quantify) and dye-based angiography (which requires an invasive procedure). Optical coherence tomography angiography (OCTA) provides several advantages relative to these allied imaging modalities, including non-invasive depth- resolved imaging at capillary-scale resolution. These capabilities imply the potential to provide DR diagnosis and staging with a single procedure. Research at Oregon Health and Science University (OHSU) demonstrated that quantification of pathological features in OCTA data is able to accurately diagnose and stage DR. Artificial intelligence has the potential to make this process more reliable than the current standard of care. In this proposal, a partnership between IFOCUS IMAGING LLC and OHSU, we seek to bring AI-aided DR diagnostics based on OCTA to the bedside by developing a software platform, OCT-ART-DR, that will automate the key components of OCTA image grading. This will be accomplished through two specific aims. First, we will improve AI-based segmentation of features in OCTA images relevant to DR. We will target three segmentation tasks: (1) segmentation of non-perfusion area, which is a key vascular pathologic feature in early DR development and can be used to stage the disease through intermediate severities; (2) segmentation of macular retinal fluid, which accumulating evidence suggests may be the most sensitive indicator of vision-threatening diabetic macular edema (the leading cause of vision loss in DR); and (3) retinal neovascularization area, which can cause catastrophic vision loss. In the second aim, we will develop an AI-based end-to-end interpretable automatic DR diagnosis framework. This will be accomplished using state-of-the-art classification networks separately trained for three goals, in ascending specificity: diagnosis of DR, diagnosis of referable DR, and diagnosis of vision-threatening DR. Both of these capabilities- pathology segmentation and interpretable end-to-end diagnostics- will be combined in the OCT-ART-DR platform developed by IFOCUS. This platform will be designed for seamless integration into several contexts, including the clinic, reading centers, and research labs. We believe that the advantages provided by our platform will lead to its voluntary adoption by both clinicians and researchers. We also expect that the introduction of this technology could be a driver for OCTA imaging to replace legacy imaging modalities, which would lead to improved DR management and, ultimately, better patient outcomes.