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

Eye diseases are a significant health concern that adversely impacts human life. Cataracts, diabetic retinopathy, and glaucoma are some of the diseases that cause irreversible and serious health problems. Eye health is greatly influenced by age, genetics, and environmental factors. Proper diagnosis of eye ailments is crucial, as it ensures accurate and effective treatment. The proximity of disease detection to error for accurate and personalized treatment intensifies the clinician's responsibility further. Developing technology and deep learning make it feasible to determine if an individual has an eye disease, and to identify the specific disease. The objective of this research is to design resolutions for detecting significant health issues such as eye diseases with the aid of deep learning models. DenseNet, EfficientNet, Xception, VGG, and ResNet architectures, which are prominent Convolutional Neural Network models, are utilized to address the issue at hand. Technical term abbreviations are explained where first used. The dataset employed for detecting diseases in retinal fundus images consists of a total of 4217 images, comprising 1038 cataracts, 1098 diabetic-retinopathy, 1007 glaucoma, and 1074 healthy individuals. The performance of the tested models was assessed using evaluation metrics such as accuracy, recall, precision, F1-score, and Matthews's correlation coefficient metrics through 10-fold cross-validation. Upon analysis of the classification performances, the EfficientNet model obtained the best results for these evaluation metrics at 87.84%, 92.84%, 94.41%, 93.53%, and 83.87%, respectively. Thus, EfficientNet architecture delivered the best classification performance in this context.