PROJECT SUMMARY Diabetic retinopathy (DR) is a leading cause of vision loss. This vision loss is largely preventable through timely diagnosis and treatment. The current method for identifying eyes at risk relies on qualitative evaluation of key features on clinical examinations or fundus photographs. While this approach stratifies risk, it does so imprecisely, requiring referral of many patients in order to identify the few who need treatment. We propose an alternative approach that focuses on the direct precursors of vision-threatening complications using advanced optical coherence tomography angiography (OCTA). This approach has the potential to more precisely identify patients at high risk for these complications than photographic screening, reducing the burden on the system. Our preliminary data has identified OCTA precursors of the vision-threatening DR complications: proliferative diabetic retinopathy (PDR) and center-involved diabetic macular edema (CI-DME). To use these biomarkers in real clinical settings, OCTA must have a wide field of view with high resolution while also providing a high imaging yield. In addition, these biomarkers must be accessible to clinicians, requiring automated and reliable identification and segmentation. The currently available systems fall short of these requirements. We propose making fundamental technological improvements to address the present limitations and to test our hypothesis that the OCTA precursors of PDR and CI-DME can precisely identify eyes at risk: In Specific Aim 1, we will develop a high-speed (1-MHz), wide-field (120?), and high-resolution (10-µmtransverse resolution) system capable of reliable, high-yield OCTA from a single scan that does not require montaging or eccentric fixation. This field of view is comparable to currently available ultra-widefield fluorescein angiography. GPU-based real-time processing will reduce motion artifacts and improve scan acquisition reliability. In Specific Aim 2, we will develop deep-learning algorithms that identify these precursors. The precursors of neovascularization (pNV) are epiretinal hyperreflective material with flow signal above the internal limiting membrane that are invisible with conventional imaging. Active microaneurysms (aMA) are characterized by hyperreflective walls with flow signal within the lumen. We will use deep-learning techniques to accurately segment and quantify these features and present the biomarkers in intuitive and interpretable format in real time. In Specific Aim 3, we will evaluate the clinical significance of advanced-OCTA derived biomarkers. (1) In a 2- year prospective study, we will follow 100 patients with moderate to severe nonproliferative DR at baseline for development of PDR or CI-DME. (2) In a cross-sectional study, we will compare the sensitivity for detecting neovascular lesions in 50 eyes with severe NPDR and PDR using the advanced OCTA vs. ultra-widefield fluorescein angiography vs. fundus photographs. (3) In...