A Review on Detection of Diabetic Retinopathy

Abstract

AbstractDiabetic retinopathy (DR) is a leading cause of diabetes mellitus, which seeds lesions on the retina that helps in sequel vision. If it is not identified in the beginning, it will lead to blindness. Unfortunately, DR is an unalterable process and treatment only assists vision. The risk of vision loss has a remarkable reduction with early detection and treatment. The process of manual detection of fundus DR images by ophthalmologists is time-consuming, cost-intensive, and vulnerable to misdiagnosis rather than computer-aided diagnosis methods. Lately, deep learning is one of the most familiar techniques that has attained finer performance in numerous fields, mainly in medical image scanning and categorization. The objective of this survey paper is to provide a review of various methods applied to diabetic retinopathy. A comparative study is made and the highest accuracy achieved is 92.5% for the detection of DR using a convolutional neural network (CNN). In this survey paper, the recent state-of-procedures for diabetic retinopathy color fundus image identification and categorization using DL strategies have been evaluated and analyzed. In addition, the DR datasets for the color fundus images of the retina have been revised. Divergent challenging issues that need detailed examination are also analyzed.KeywordsDeep learning (DL)Diabetic retinopathy (DR)Convolutional neuralNetwork (CNN)LesionImage processingDiabetic macular edema