<![CDATA[Providing quality healthcare is a fundamental right of citizens as stipulated in the 1945 Constitution, making healthcare a national priority as outlined in the Ministry of Health's 2020-2024 Strategic Plan. However, high patient visitation rates can lead to overcrowding, impacting service efficiency and quality. Therefore, real-time patient monitoring technology is needed. Previous studies have shown promising results, but remain limited to ideal conditions for the machine. This study uses the YOLO algorithm to detect patient congestion in real healthcare facilities using CCTV footage from waiting rooms. This study uses three instance segmentation models —YOLOv8n-seg, YOLO11n-seg, and YOLOE-seg —that are tested on a custom dataset and compared with the official model. The results of training the custom dataset model are: YOLOv8n-seg Precision 96%, Recall 97%, mAP50 98%, mAP50-95 84%, and F1-score 97%. YOLO11n-seg precision 96%, Recall 97%, mAP50 98%, mAP50-95 84%, and F1-score 97%. and YOLOE-seg precision 96%, Recall 98%, mAP50 98%, mAP50-95 85%, and F1-score 97%. In addition, this study compared predictions with the official model, which found that all custom dataset models successfully detected objects with 100% density. In contrast, the official model correctly predicted density 70%-82% of the time. Therefore, this study concludes that models trained on custom datasets can improve the accuracy of patient density predictions, thereby enhancing the quality of real-time healthcare services.]]>
Copyrights © 2025
