Performance evaluation for classifying type 2 diabetic retinopathy using deep neural network

Now-a-days irrespective of age and gender, most people are being affected by retinal diseases. People with type 2 diabetes are more prone to blindness. Periodic check-up for diabetic retinopathy (DR) has become labour intensive task. Even though many methods based on computational intelligence are proposed for detecting diabetic retinopathy, those methods are not efficient in classifying DR type. Early diagnosis and proper follow up treatment can prevent progressing to next stages of DR. This paper presents an automatic disease detection that utilises retinal image analysis to accurately categorise the retinal problem as normal, non-proliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR). This system uses three steps to analyse fundus images and to classify the severity grade using deep neural networks. Test results showed that our proposed system could classify the DR with 96.3 of accuracy for SVM, 95.2 accuracy for k-NN, 99.15 for ANN, and CNN scored an accuracy of 0.7998 and loss of 0.4569. ANN proved to be better when compared to existing works. Different k-values are taken for k-NN and when k = 5 accuracy is 95.2.