Multi-Class of Retinal Diseases Classification via Deep Learning Techniques Based on Fundus Images

Abstract

Early detection of retinal disorders is highly important in making the decision at the right moment because of the importance of treatment and prevention of diseases that are a significant cause of blindness. Experts in diagnosing these diseases frequently encounter a number of obstacles, including time, as well as variances across observers because of the occurrence of a comparable set of symptoms among groups, and some may share Categories in similar instances to address these challenges. This work proposes a novel model of deep learning approaches concerning object detection and classification. By utilizing YOLOv8, YOLOv8m-cls was used, which is distinguished by excellent efficiency, accuracy, and speed and outperforms earlier investigations. The model was trained using publicly available data from the Kaggle database, which contains 4217 fundus images, including glaucoma, cataracts, diabetic retinopathy, and the final health condition. The accuracy was 94, the precision was 95%, the recall was 97%, and the F1-score was 96 after 0.5 hours of training. This study demonstrates, based on the data obtained, particularly with the new model YOLOv8m-clsIn, revolutionizing the diagnosis of eye diseases, which leads to better patient outcomes and health care provision.