Automatic Segmentation and Machine Learning Classification Using Optical Coherence Tomography Angiography

Abstract

Diabetic retinopathy emerges as a consequence of untreated chronic diabetes, posing a risk of total blindness if not promptly addressed. Early diagnosis and treatment are pivotal in averting the severe consequences associated with diabetic retinopathy. Traditional identification of diabetic retinopathy by an ophthalmologist is time-consuming, subjecting patients to prolonged discomfort. Introducing an automated method could facilitate immediate diagnosis, allowing for convenient follow-up therapy to prevent potential eye damage. This study suggests using machine learning to extract three important features: microaneurysms, hemorrhages, and exudates. A hybrid classifier is used for the classification, which combines machine learning classifier algorithm and neural network. The results of the study indicate that the hybrid strategy can achieve up to 82% maximum accuracy, with corresponding precision, recall, and f1 scores of 81 %, 81.2%, 80.3%.