Clinical Features Distinguishing Diabetic Retinopathy Severity Using Artificial Intelligence

Background and Hypothesis:1 in 29 American diabetics suffer from diabetic retinopathy (DR), the weakening of blood vessels retina. DR goes undetected nearly 50% diabetics, allowing to steal vision many Americans. We hypothesize that increasing rate ease diagnosing by introducing artificial intelligence-based methods primary medical clinics will increase long-term preservation ocular health patients.
 Project Methods:This retrospective cohort study was conducted under approval Institutional Review Board Indiana University School Medicine. Images were deidentified no consent taken due nature this study. categorized 676 patient files based upon HbA1c, severity non-proliferative (NPDR), proliferative (PDR). Retinal images annotated identify common features DR: microaneurysms, hemorrhages, cotton wool spots, exudates, neovascularization. The VGG Image Annotator application used for annotations allowed us save structure coordinates into a separate database future training intelligence system.
 Results:228 (33.7%) patients diagnosed with diabetes, 143 (62.7%) those DR. Two-sample t tests found significant differences between HbA1c values all compared without (p<0.007) severities versus (p<0.002). 283 eyes form study: 37 mild NPDR, 42 moderate 56 severe 148 PDR eyes.
 Potential Impact:With dataset experiment, we aim train an system diagnose through retinal imaging. goal is be conveniently detection preserve millions