Image Analysis of Diabetic Retinopathy Disease Based on Artificial Neural Network Algorithms

Abstract

Diabetic retinopathy is a complication of diabetes in the form of damage to the retina of the eye. High levels of glucose in the blood are the cause of small capillary blood vessels to rupture and can cause blindness. The signs of this disease can only be seen using retinal fundus images. To identify diabetic retinopathy, a computerized process and analysis are needed, one of which uses artificial neural network methods to determine its performance so that it will help the doctor in analyzing the disease and diagnosing whether a patient suffering from diabetic retinopathy or not. Texture feature extraction method using Gabor filter can represent feature value information that is skewness, kurtosis, mean, entrophy, and variance to be processed at the identification stage using artificial neural network methods. The comparison results of the DIARETDB0 dataset testing with the total of 130 fundus images using the backpropagation ANN method before randomizing the data yielded an accuracy value of 82.30%, a precision value of 71.28%, a recall value of 82.30%, and an f-measure of 76.39%. Whereas after randomizing the data for 30 times, the results of accuracy value were higher than before randomizing the data, namely the accuracy value of 83.07%, the precision value of 71.39%, the recall value of 83.07% and f-measure value of 76.78%. The tests carried out included in good classification.