Classification of Fundus Images For Diabetic Retinopathy using Artificial Neural Network

Abstract

People with diabetes may suffer from an eye disease called Diabetic Retinopathy (DR). This is caused by damage to the blood vessels of the light-sensitive tissue at the back of the eye (i.e retina). Fundus images obtained from fundus camera are often imperfect; normally are in low contrast and blurry. Hence, causing difficulty in accurately classifying diabetic retinopathy disease. This study focuses on classification of fundus image that contains with or without signs of DR and utilizes artificial neural network (NN) namely Multi-layered Perceptron (MLP) trained by Levenberg-Marquardt (LM) and Bayesian Regularization (BR) to classify the data. Nineteen features have been extracted from fundus image and used as neural network inputs for the classification. For analysis, evaluation were made using different number of hidden nodes. It is learned that MLP trained with BR provides a better classification performance with 72.11% (training) and 67.47% (testing) as compared to the use of LM. Such a finding indicates the possibility of utilizing BR for other artificial neural network model.