Comparative Analysis of YOLOv5 and YOLOv8 for Tear Film Lipid Layer Detection: Architectural Disparities, Performance Metrics, and Future Implications

This study employs the YOLOv5 and hypothetical YOLOv8 algorithms to analyze 52 real-world ocular surface videos for tear film lipid layer detection. A comprehensive evaluation of architectural disparities, merits, and demerits is conducted, emphasizing critical measurement parameters, including mean Average Precision (mAP), Specificity, recall, fl-score, and Frame Per Second (FPS). The assessments take place on a PC featuring an Intel Core i5-10500H CPU, with an input size of 640*640 pixels. Results indicate significant performance distinctions, with YOLOv8 demonstrating superior accuracy at a potential cost to computational efficiency. A proposed future research direction explores dry eye detection based on liquid detection using natural images and videos captured by mobile phones, aiming to leverage widespread accessibility for enhanced diagnostics in real-world settings.