Identification of Severity Level for Diabetic Retinopathy Detection Using Neural Networks

The diabetic retinopathy is a vital factor of vision loss among individuals with diabetes. Early detection of this condition has significant importance for the patient’s vision. In this proposed work, neural network architectures are implemented by considering the existing methods as base methods and developed a model which is unique for detection of diabetic retinopathy among diabetic patients by screening the fundus images through the proposed models. Timely detection and appropriate treatment can help to prevent the beginning and development of diabetic retinopathy among diabetic patients. Accurate detection of the disease is an essential requirement in the health domain. Our focus of the research is to classify the severity level using multiclass classification for diabetic retinopathy. The classification results, executed on the Google Colaboratory platform, indicated that CNN, VGG16 and GoogleNet architectures yielded accuracies of 73.44%, 75.25% and 73.93%, respectively.