"A 30-Year Aspiration in Sleep Medicine": AI Sleep Test Model Trained on Household Noise Achieves Hospital-Level Sleep Study Accuracy

Asleep, a sleeptech company, in collaboration with the research team led by Dr. Yoon In-Young from the Department of Neuropsychiatry at Seoul National University Bundang hospital, has developed an AI model that has learned various sounds, including household noise. This innovation enables the accurate measurement of sleep stages not only in a hospital setting but also in typical households.

This research result is not only scheduled for publication in ‘JMIR(Journal of Medical Internet Research)’, prestigious international journal in the field of health informatics, but has also been introduced at SLEEP 2023, an academic conference hosted by the American Academy of Sleep Medicine(AASM), held in Indianapolis, USA, since the 3rd of last month.

According to the joint research team, this study aimed to implement AI models that were originally trained in a hospital setting into a household environment. To achieve this, the AI model was trained to recognize various sounds that occur when people sleep at home. The research team said, "We used 6,600 hours of sound data recorded with smartphones during sleep in a household environment, as well as 270 hours of breath sound data recorded through smartphones during household sleep studies."

As a result, when the AI model trained on polysomnography data from a hospital environment was applied to a home environment, it performed about 85% of the results measured in the hospital, but the AI model trained on sound data from the home environment performed about 10% better than that.

Household polysomnography are generally known to offer a higher accuracy in monitoring sleep stages compared to hospital-based polysomnography because they take place in a real sleep environment. However, they also have drawbacks, including limitations in human resources and supervision during the tests. Dr. Kim Dae-woo, Head of AI at Asleep and a participant in the research, emphasized, "This study represents the world's first sound-based sleep stage research conducted in a household environment," and added, "It demonstrated that it maintains a high level of accuracy even in household settings compared to conventional hospital-based AI models."

The joint research team further highlighted that "this research result has received international recognition, not only through publication in international journals and the American Academy of Sleep Medicine(AASM) but also through its presentation at 'ICLR 2023,' the world's most prestigious conference in the field of AI."

[Related Articles]

Medigate - Asleep Unveils Sound-Based Sleep Stage Measurement Deep Learning Model at the American Academy of Sleep Medicine (June 4, 2023)

Maeil Business - Asleep AI Sleep Model Ushers in the Era of In-Home Sleep Testing (June 6, 2023)