Classification of diabetic retinopathy stages using image segmentation and an artificial neural network

Abstract

Diabetic retinopathy, which can lead to blindness, has been found in 22% of diabetic patients in the latest survey. Therefore, diabetic patients should have an eye examination at least once a year. However, it has been found that currently there is a problematic lack of specialists in ophthalmology. Detection and treatment of diabetic retinopathy are thus delayed. The idea to create a classification system of diabetic retinopathy stages to facilitate the making of preliminary decisions by ophthalmologists is introduced. This paper presents the classification of diabetic retinopathy stages using image segmentation and an artificial neural network. This proposed method applies local thresholding to separate the foreground region from the background region so that the optic disc and exudates regions are able to be identified more clearly. The experiment was carried out with 100 fundus images from the Institute of Medical Research and Technology Assessment database. The prediction model had an accuracy rate of up to 96%.