Smartphone AI: Personalized Precision Sleep Analysis

Sleep-tech company ASLEEP, in collaboration with Professor Yoon In-young of the Department of Psychiatry and Professor Kim Jung-hoon of the Department of Otorhinolaryngology at Seoul National University Bundang Hospital, has successfully demonstrated the performance of an AI model that accurately distinguishes sleep stages by separating individual breathing sounds, even in a multi-person sleep environment. This is lauded for having opened new horizons in sleep analysis technology by resolving the accuracy problem in multi-person environments, which was previously considered the biggest limitation of existing sleep analysis technologies. The research team created a realistic environment where 44 adult pairs (total 88 people) slept simultaneously in one bed. They then placed smartphones next to each person's pillow to record breathing sounds, while simultaneously conducting polysomnography (PSG) to compare and verify the AI model's performance. As a result, the ASLEEP AI model recorded a Macro F1 score of 0.63 in 4-stage sleep classification (awake, REM sleep, light sleep, deep sleep) and 0.77 in 2-stage classification (awake/sleep), showing high predictive accuracy comparable to polysomnography. This represents an improvement of approximately 29% compared to the 4-stage standard performance (Macro F1 score of 0.49) of existing wearable sleep trackers. This study demonstrated that precise analysis comparable to polysomnography is possible with just a smartphone, suggesting a way for anyone to easily manage their sleep health without wearable devices. Furthermore, it is the first case to clinically prove applicability in a shared sleep environment, laying the groundwork for future research to diagnose and monitor sleep-related disorders such as snoring and sleep apnea in multi-person environments. The algorithm used in the study is already integrated into products of major partners such as SKT A.dot and Samsung Life's The Health, including 'Abnotrack', a Ministry of Food and Drug Safety Class 2 medical device for diagnosing sleep apnea, thus recognizing its technological prowess. The results of this study were published in the international journal of sleep medicine, 『Sleep Medicine』, and achieved the feat of being selected as an outstanding abstract at the 2024 European Sleep Research Society (ESRS).