Advancing Medical Diagnostics: Performance Analysis of YOLO Algorithms for Automated Platelet Detection

The significance of health automation, particularly post-COVID-19, emphasizes the critical role of advanced diagnostic technologies. Conventional methods for platelet detection in blood-smear images are often slow and inconsistent, impacting medical response efficiency. This study explores the application of the YOLO (You Only Look Once) algorithm, specifically versions 3, 5, 7, and 8, to improve platelet detection accuracy and efficiency. Through meticulous data preparation and fine-tuning on a curated dataset, we compared the performance of these YOLO versions in terms of accuracy, precision, recall, and computational efficiency. YOLOv5 emerged as the most accurate and reliable model, achieving the highest mean Average Precision (mAP) and superior precision and recall metrics, despite slightly slower inference times compared to YOLOv8. YOLOv8, with the fastest inference times and excellent detection capabilities, is also highly effective for real-time applications. The findings underscore the advancements and tradeoffs among different YOLO versions, with YOLOv5 and YOLOv8 showing significant potential for enhancing automated platelet detection in clinical diagnostics.