Sleep apnea is a disorder that causes people to stop breathing multiple times during their sleep, when untreated. It can be diagnosed trough polysomnography (PSG), which is a time consuming, expensive and must be performed in special laboratories. Due to its complexity, different alternatives to PSG have been developed. This paper presents a system based on the edge-computing paradigm for detection and alerting of sleep apnea events, using data from a single-channel ECG sensor. A framework for automated feature selection is used for the extraction and selection of the important features. Some ECG signal specific features were also added to the generic framework. We have evaluated several machine learning algorithms for sleep apnea detection based on the generic features and the ECG-specific features on a dataset containing 70 recordings, available in the PhysioNet database. The obtained results show that the combination of generic features and ECG-specific features improve the detection accuracy to up to 82% with a small set of about 20 computationally efficient features.
Sleep apnea PhysioNet ECG QRS Feature extraction