Diagnosis of Diabetic Retinopathy Using Deep Learning Techniques

Abstract

Diabetic retinopathy (DR) is a critical eye malady and a severe cause of visual deficiency in diabetic patients. Diabetic retinopathy is the root cause of more than 1% of the visual deficiency around the world. DR affects blood vessels of retina which may lead to diabetic macular edema, neovascular glaucoma and retinal detachment. To prevent DR from progressing and causing severe damage, an early detection of diabetic retinopathy is crucial. The analysis of diabetic retinopathy (DR) through color fundus images needs skilled clinicians to discover the presence of DR. The grading system is also very hard. Overall the process is less optimal and very time consuming. In this paper, we propose different deep learning approaches to diagnose DR from digital retinal fundus images and precisely classify its severity. In order to build the best classification, simulate the various deep learning models such as ResNet50, ResNet152V2, VGG16 and VGG19. Different benchmark datasets with different characterizations and complexities have been used for training and testing the models. The simulation results illustrate that the proposed method gives better performance while comparing with other methods.