Deep Learning-Based Classification of Ocular Diseases Using Convolutional Neural Networks

Abstract

Ocular diseases can range from mild discomfort to severe vision loss or even blindness, and can affect people of all ages. Early detection and treatment are crucial to prevent or delay vision loss and maintain eye health, especially for older adults or those with underlying medical conditions. Early detection and rapid treatment of ocular problems can be done using Computer vision / Deep learning tasks. The proposed method must be able to distinguish between six distinct diseases, including glaucoma, cataract, diabetes, age-related macular degeneration, hypertension, and pathological myopia, as well as other diseases that are not specifically mentioned, in the Ophthalmic Disease Recognition (ODIR) dataset. Due to the large degree of variation in picture quality, disease presentation, and patient demographics, the ODIR dataset presents a difficult job for multiple classification, so the accuracy was below 61%. In this study, the ODIR dataset is used to improve the suggested model, and perform extensive experiments to optimize the hyperparameters for training. The findings show that the suggested approach successfully completes the binary classification task on the ODIR dataset with excellent accuracy between 98% and 100%, recall from 97.99% to 100%, and precision between 96% and 100%.