SYSTEM FOR DETECTING IN-BED PATIENT POSITION AND NURSE ALERTING FOR PRESSURE ULCER PREVENTION USING DEEP LEARNING AND FLASK WEB APPLICATION

Sleep posture analysis is essential for addressing conditions such as sleep apnea, pressure ulcers, and other posture-related complications. Proper monitoring and timely intervention can significantly improve patient outcomes by preventing these issues. This study presents a robust deep learning model, YOLOv8n, designed to accurately detect patient positions in bed, classified into four key postures: Supine, Prone, To-Left, and To-Right. Trained on a dataset of 4514 images — 3160 for training, 903 for validation, and 451 for testing — the YOLOv8n model achieved exceptional performance, with a precision of 95.2%, a recall of 88.2%, and an [Formula: see text]1-score of 92% during testing. To further assess its robustness, an external dataset of 150 images (100 for validation and 50 for testing) was used, where the model demonstrated strong generalization capabilities, achieving a precision of 83.3%, a recall of 93.5%, and an [email protected] of 93.6%. In the final testing phase, YOLOv8n demonstrated superior performance compared to YOLOv5n, achieving metrics such as precision (95.2% versus 88.2%), recall (88.2% versus 73%), [email protected] (93.7% versus 80.7%), and [email protected]:0.95 (68.6% versus 54.0%). This 15.4% improvement in [email protected]:0.95 underscores YOLOv8n’s superior accuracy, particularly in challenging postures like Prone and To-Right. These results confirm its reliability and robustness for real-time patient posture detection in diverse conditions. To enhance its practicality, the YOLOv8n model was integrated into a web application developed with the Flask micro web framework. The application provides real-time position detection and features an alert system to notify caregivers when a patient remains in the same posture for more than the medically recommended duration (1–2 h). This integration ensures usability in both clinical and home healthcare environments. In conclusion, YOLOv8n offers a highly accurate, reliable, and practical solution for real-time patient posture monitoring. By outperforming YOLOv5n across all metrics, YOLOv8n emerges as a superior model, well-suited for diverse healthcare applications and capable of improving patient care through timely intervention and effective monitoring.