Deep Learning Based Multi-Class Eye Disease Classification: Enhancing Vision Health Diagnosis

Abstract

Retinal abnormalities impact millions of people globally. Timely detection and treatment of these abnormalities could prevent further progression, potentially saving countless individuals from preventable blindness. However, manual disease detection is a slow, laborious process and lacks consistency in results. This study uses convolutional neural networks to categorize eye disease using a publicly available dataset. Five different pre-trained models based on convolutional neural networks (CNNs), including VGG-16, VGG-19, Resnet-50, Resnet-152, and DenseNet-121, were used in this study. We were able to detect eye diseases at the cutting edge using the refined VGG-19. With testing accuracy of 95% on the dataset, this model accurately predicted eye diseases due to the effective and same weighted precision, recall, and F1 score of 95%. The model also significantly reduces training loss while improving accuracy.