A Comparison of YOLOv5 and YOLOv8 in the Context of Tear Film Break-Up Detection Based on Ophthalmic Videos

This research study presents a comparative assessment of two prominent deep learning models, YOLOv5 and YOLOv8, in the specific context of tear film break-up area detection within ophthalmic videos. Performance metrics, including mean Average Precision (mAP), sensitivity, recall, F1-score, and Frames Per Second (FPS), were measured for YOLOv5 and YOLOv8 using a PC with an Intel Core i5-10500H CPU, with an input size of 640x640 pixels. The results indicate that YOLOv8 outperforms YOLOv5 across various metrics, exhibiting superior mAP, sensitivity, recall, and F1-score, while also achieving a higher FPS. This study holds significant importance as it offers valuable insights for the application of YOLO-based deep learning algorithms in the field of ophthalmology and other medical domains. Moreover, comprehending the distinctions in model architecture between these two models holds notable importance in the context of augmenting model performance through the application of explainable artificial intelligence techniques. Besides, the team develop a user-friendly TBUT detection system of “AI Dry Eye Analytic System”, which is available at”mini.ac.cn”.