Automated sleep apnea detection from ECG signals based on flexible analytic wavelet transform

Abstract

Apnea considered as a serious sleep issue. It occurs when a person’s breathing is disrupted due to not having adequate oxygen supply to brain and also body while sleeping. Severe case in this sleep apnea is obstructive sleep apnea (OSA). The OSA detection at early stages is necessary to improve the life quality. So to aid in computer diagnostics and have better automated identification in home this method is being proposed. This method proposes the automated classification of ECG signals affected OSA employing a flexible analytic wavelet transform (FAWT) based approach. The log energy (LOEN), approximate entropy (APEN), and Hurst exponent (HUEX) features have been employed. The 35-folded cross validation technique was employed using different classifiers like K nearest neighbor (KNN), support vector machine (SVM), linear discriminant (LD), logistic regression, complex tree to classify into normal and apnea affected subjects. This proposed method has achieved highest accuracy as 90.3% with area under curve (AUC) 0.92. Our model has shown promising results developed using the publicly available database.