Classification of diabetic retinopathy and normal fundus images based on texture features using Multilayer Perceptron (MLP)

Abstract

Diabetic retinopathy is a disease caused by uncontrolled blood sugar levels and occurs continuously. Funduscopic examination with an ophthalmoscope tool to determine diabetic retinopathy. This study aims to classify funduscopy images in distinguishing normal eyes and diabetic retinopathy based on texture characteristics using the multilayer perceptron (MLP) method. Texture feature extraction as a class recognition process that aims to produce characteristics based on the texture of each image. The texture features used are histogram and GLCM with 10 parameters. Research data is sourced from the Zenodo website and the National Library of Medicine. Based on the results of the study, it shows that the multilayer perceptron method with the help of Weka machine learning in classifying eye fundus images to distinguish normal eye cases and diabetic retinopathy produces an accuracy value of 83.75% at k-folds 20 cross validation with sensitivity and specificity values of 49.20% and 95.09%.