Vision transformer for detecting ocular diseases.

Ocular disease detection is critical for effective treatment and preventing potential vision loss. This study aims to accurately detect different ocular diseases, including cataracts, foreign bodies, glaucoma, subconjunctival hemorrhage, and viral conjunctivitis, based on a new deep learning architecture called "vision transformer (ViT). The proposed imaging system is experimented on a dataset of 500 eye images collected from many patients in ophthalmic hospitals in addition to 100 images of healthy participants. The performance of ViT model with different batch sizes (5, 10, 15, and 20) was evaluated using different metrics, including confusion matrix, accuracy, precision, recall, F1-score, G-score, and zero-one loss to identify which model achieved the best performance. The experiments on the collected data demonstrate that the proposed imaging system can effectively detect ocular diseases and achieve competitive performance, thus providing additional insights into the model's decision-making process. [ABSTRACT FROM AUTHOR]