Deep-learning based sleep apnea detection using sleep sound, SpO2, and pulse rate

Sleep apnea, a common sleep disorder where breathing is repeatedly interrupted during sleep, poses significant health risks. Traditional diagnostic methods like overnight polysomnography (PSG) are complex and costly, limiting widespread screening. This study suggests a deep learning model to identify sleep apnea using sleep sounds, oxygen saturation (SpO2), and pulse rate. Mel-spectrogram, computed from PSG data of 24 patients, serves as input. The study shows the effectiveness of deep learning, with the combined model achieving 96% accuracy in inferring apnea severity, outperforming individual models using SpO2 and pulse rate (79%) and sleep sound (83%).