Computerized obstructive sleep apnea diagnosis from single-lead ECG signals using dual-tree complex wavelet transform

An algorithm for apnea identification using one-lead electrocardiogram is presented in this article. Segments of ECG signals are fed into dual-tree complex wavelet transform (DT-CWT) to generate frequency sub-bands. Three statistical moment parameters are then developed from the DT-CWT outputs. The suitability of these statistical moments in distinguishing normal and apneic ECG signals is investigated through extensive analyses. The overall algorithmic detection accuracy of the scheme is determined for various machine learning classifiers. Sleep apnea classification is done using logistic boosting (LogitBoost). Until now this is for the first time LogitBoost has been implemented for automated sleep apnea detection. We also performed cross validation for classification model evaluation and optimal parameter selection. Results suggest that the detection accuracy of the apnea screening scheme presented in this work is comparable to extant sleep apnea screening systems of the literature. It can be anticipated that upon its implementation, the detection scheme proposed in this work will take us one step closer to sleep apnea monitoring device implementation and eradicate the onus of the physicians.