Development and Clinical Validation of Artificial Intelligence Algorithms for Sleep Staging and Obstructive Sleep Apnea Screening Based on a Large-Scale Model Trained on Multimodal Physiological Signals From Diverse Populations

In recent years, wearable and contactless devices, including smart rings, smart bands/watches, and smart mattresses, have developed rapidly and can continuously collect multidimensional data such as heart rate, oxygen saturation, respiration, body movement, snoring, body temperature, and environmental parameters. Nevertheless, most current consumer-grade or wellness-oriented devices have not been trained and validated using large polysomnography (PSG)-synchronized reference datasets. Their accuracy, robustness, and interpretability in people with insomnia, obstructive sleep apnea (OSA), and other sleep disorders therefore require systematic evaluation. Leveraging a multicenter clinical sleep-medicine platform, this study will synchronously collect standard full-night polysomnography (PSG) data, continuous body-temperature data, smart-ring sensor data, contactless sleep-monitoring belt data, and indoor environmental monitoring data from the same participant during the same monitoring night. A clinical sleep database will be established with PSG-derived reference-standard labels, strict temporal synchronization, and multimodal features. The database will support the development, optimization, and validation of sleep-staging and obstructive sleep apnea (OSA)-screening models and will provide a data and modeling foundation for subsequent smart sleep-monitoring products, home-based sleep-health management, and clinically assisted screening applications.