A Wearable AI-Driven System for Real-Time Detection of Sleep Apnoea

Abstract

The prevalence of obstructive sleep apnoea is increasing globally, posing significant threats to sleep quality and long-term health. However, conventional diagnostic methods such as polysomnography require patients to undergo testing in sleep laboratories while wearing multiple sensors, which not only increases financial and logistical burdens but also disrupts natural sleep patterns. Furthermore, traditional approaches are not suitable for remote or real-time monitoring, and existing studies primarily focus on post-hoc data analysis rather than live detection and intervention. To address these limitations, this study proposes a wearable, AI-driven system for real-time detection and management of sleep apnoea. The system integrates multiple sensors—including tri-axial accelerometers, gyroscopes, PPG, and microphones—to continuously monitor respiratory activity, heart rate, blood oxygen saturation, and body posture. A deep learning model based on the YOLOv9 architecture is implemented for event-level detection of apnoea and respiratory events. The trained model achieved an apnoea detection precision of 95.7%, a mean average precision (mAP@0.5) of 94.6%, a recall of 85.7%, and an overall classification accuracy of 89% based on the confusion matrix. Validation against the SOMNOTouch device confirmed the reliability of the system, with apnoea detection accuracy reaching 93% using predictive time-window enhancement. The real-time detection engine demonstrated a response latency of 118.29 ± 11.68 ms, enabling timely visual feedback and potential activations of therapeutic stimulators. The system architecture includes AI model deployment, mobile application development, and cloud-based storage infrastructure to support continuous monitoring, model updates, and remote analysis. Overall, this research contributes a scalable, cost-effective, and user-friendly solution for non-invasive, real-time sleep apnoea monitoring and intervention, bridging the gap between clinical diagnostics and home-based sleep health management.