Severity Classification of Diabetic Retinopathy using ShuffleNet

Abstract

In India, the number of people living with diabetes is about 72.96 million and large proportion of people suffer from Diabetic Retinopathy which is caused due to elevated level of blood sugar in a diabetic patient that can harm the retina of person and may lead to permanent blindness. The detection of Diabetic Retinopathy is being practiced in such a way that it still needs complex human efforts. The need of the hour is to discover new and better methods that can help further improve identification and treatment of DR (Diabetic Retinopathy). This paper proposes a different method and model to identify and distinguish DR into various severity levels. A novel approach is introduced to detect Diabetic Retinopathy using Shufflenetv2 which is Convolutional neural network. The results are better when compared to other models in literature. Smooth L2 loss function has been used to depict the errors and analyze the results.