A Novel Review of Obstructive Sleep Apnea Detection Using Photoplethysmography

Obstructive Sleep Apnea is one of the most common sleep disorders characterized by repeated airway obstruction. This causes desaturation of oxygen. The work here presented an efficient way to detect OSA based on PPG signals along with SpO2 measurement using advanced machine learning models. We compared the performance of two algorithms: LSTM and TCN against a baseline classifier on a labeled dataset. The PPG and SpO2 signals were processed, cleaned, and standardized from a public dataset before classification. In preliminary tests, we used the LSTM model for its ability to model sequences with high accuracy in detecting apnea events compared to the baseline classifier. We combined the LSTM and baseline predictions using majority voting to improve classification accuracy and robustness. Moreover, we introduced the Sleep Apnea Index (SAI) in terms of the apneic events per sample as a measure for classifying the severity of the condition. Our experiments confirmed that deep learning-based architectures for PPG signals were non-invasive and feasible ways toward OSA screening. That might have had significant effects for portable sleep monitoring devices.