Abstract
Sleep apnea is one of the most common sleep disorders, which, if left untreated, may have severe health consequences in the long term. Many sleep apnea patients remain non-diagnosed due to lacking access to medical tests. In recent years, portable and wearable sensors that measure blood oxygen saturation (SpO(2)) are becoming common and affordable for daily use, and they open the door for affordable and accessible sleep apnea screening in the context of everyday life. To learn about the advancement in SpO(2)-based sleep apnea screening, we conducted a survey of published studies. We searched databases including Springer, Science Direct, Web of Science, ACM Digital Library, and IEEE Xplore using the keywords "sleep apnea" AND ("SpO2" OR "blood oxygen saturation") AND ("machine learning" OR "deep learning"). After screening 835 results, we included 31 publications for a full-text review. Analysis shows that SpO(2)-based sleep apnea screening studies consist of three main categories: (1) individual apnea events detection, (2) apnea-hypopnea index prediction, and (3) apnea severity classification. We found two significant research gaps: a lack of sufficient and diverse publicly available datasets, and the absence of standardized protocols for data collection, signal preprocessing, and model bench marking. Future research should focus on addressing these gaps to enhance the effectiveness and reliability of AI-driven sleep apnea screening methods using SpO(2) signals.