DeepWave: Non-contact Acoustic Receiver Powered by Deep Learning to Detect Sleep Apnea

Sleep apnea impacts one billion people worldwide and if left untreated, increases risk for stroke by two times, and risk for cardiovascular mortality by five times. Traditional sleep measurement approaches are either too subjective due to relying on patients' retrospective report, or too expensive and inconvenient due to the bulky equipment like Polysomnography that requires an overnight stay in a sleep lab different from the patients' home sleep environment. To enable robust, long-term at-home daily sleep apnea monitoring, in this study we propose a non-contact acoustic receiver powered by artificial intelligence (AI) innovations named DeepWave. More specifically, we have customized an environmental acoustic receiver for remote, off-bed breath sound capturing, avoiding the need to wear devices on the body and the sensitivity of on-bed devices to posture variances. We have further developed a novel convolutional neural network to detect and discriminate even subtle changes in respiratory events even when the acoustic signal is weak. Evaluated through human experiments, DeepWave, powered by the customized acoustic receiver and deep learning, shows promising potential for reliable, long-term at-home sleep apnea measurement. DeepWave is expected to greatly advance big data-driven precision sleep medicine.