Efficient Detection of Multiclass Eye Diseases Using Deep Learning Models: A Comparative Study

Abstract

Eye diseases pose a significant health threat, impacting human life adversely. Conditions like cataracts, diabetic retinopathy, and glaucoma lead to irreversible and serious health issues. Age, genetics, and environmental factors play a crucial role in eye health. Accurate diagnosis is essential for effective treatment, placing a heightened responsibility on clinicians. Advanced technology and deep learning enable the detection and identification of eye diseases. This research aims to utilize prominent Convolutional Neural Network models, including DenseNet, EfficientNet, Xception, VGG, and ResNet, to detect eye diseases. Technical term abbreviations are explained, and the dataset comprises 4217 retinal fundus images, including 1038 cataracts, 1098 diabetic retinopathy, 1007 glaucoma, and 1074 healthy individuals. Model performance is evaluated through metrics like accuracy, recall, precision, F1-score, and Matthews's correlation coefficient using 10-fold cross-validation. Among the models tested, EfficientNet demonstrates the best results with 87.84% accuracy, 92.84% recall, 94.41% precision, 93.53% F1-score, and 83.87% Matthews's correlation coefficient. Consequently, EfficientNet proves to be the most effective architecture for classifying eye diseases in this study.