Deep Learning for Early Diabetic Retinopathy Detection

Diabetic retinopathy (DR)— a leading cause of blindness— is a diabetes complication whichcauses damage to retinal blood vessels. DR can be treated non-invasively if diagnosed early enough.However, early diagnosis requires a medical examination, which may not be possible in some demographicalregions. Much previous work has largely focused on determining the presence of DR and not the severity.Thus, the goal of this study was to develop a single neural network that could 1) detect presence or absence of DR2) perform early detection of DR 3) perform multi-class classification on DR severity.