Comparison of Classification Methods for Automatic Detection of Diabetic Retinopathy in Fundus Imagery

Abstract

Abstract Diabetic retinopathy (DR) is a complication of diabetes. According to an article in the online newspaper compass published 15/08/08, it is estimated that the number of diabetics which originally numbered 117 million in 2000 will increase to 366 million in 2030. In this study 94 data were used. The classification process of the DR includes four main stages, namely Preprocessing, segmentation, feature extraction and classification. The system built in this research is automatic detection of Diabetic Retinopathy in fundus images from images obtained from STARE (Structured Analysis of the Retina). This research was conducted to create an application that can display the results of classification automatically by comparing the two methods of artificial intelligence, namely SVM (Support Vector Machines) with ANN (Artificial Neural Network), the results of the comparison of the two methods of artificial intelligence that is to ANN produce accuracy values of 82%, a specification value of 80%, and a sensitization value of 84%, and an NPV value of 81%, and a PPV value of 87%. SVM produces an accuracy of 79%, a specification value of 72%, and a sensitization value of 84% and NPV value of 81% and PPV value of 86%. From this value it can be seen that the ANN artificial intelligence method is better for classification compared to SVM.   Keywords: Diabetes Mellitus, Artificial Neural Network, Support Vector Machines, PDR, ME, Feature extraction, Classification