Open AccessAutomated Recognition of Sleep Apnea-Hypopnea Syndrome Using Continuous Wavelet Transform-Based Multiscale Dispersion Entropy of Single-Lead ECG Signal
Author(s) -
Hadj Abdelkader Benghenia,
Hadj Slimane Zine-Eddine,
Andrade Alexandre
Publication year - 2022
Publication title -
traitement du signal/ts. traitement du signal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.279
H-Index - 11
eISSN - 1958-5608
pISSN - 0765-0019
DOI - 10.18280/ts.390136
Subject(s) - pattern recognition (psychology) , support vector machine , artificial intelligence , wavelet transform , approximate entropy , polysomnography , sleep apnea , heart rate variability , continuous wavelet transform , sample entropy , wavelet , speech recognition , computer science , mathematics , apnea , discrete wavelet transform , medicine , cardiology , heart rate , blood pressure
A new method is proposed for recognition of the sleep apnea-hypopnea syndrome (SAHS) using electrocardiograms (ECG) signal in order to find an alternative with the same performance of polysomnography (PSG). Heart rate variability (HRV) signals generated from an ECG signal are used to examine a wide range of indices. A novel aspect of this work is the use of a method to decompose the HRV spectrum into total power spectrum (TP), high frequency (HF), low frequency (LF) and very low frequency (VLF) sub-band signals, and correlates their energy content with sympathetic and parasympathetic activity. The HRV signal was decomposed using the continuous wavelet transform (CWT) followed by the inverse continuous wavelet transform (ICWT), and sub-band signals were extracted from 5-minute episodes. In this regard, the suggested technique provides novel indices based on the mean of small (1 to 5), medium (6 to 10) and large (11 to 20) time scales of multiscal dispersion entropy (MDE) for each sub-band signals. In order to choose the best classifier, the indices of the MDE are submitted to a t-test technique and categorized using three classifiers: decision trees (DT), support vector machines (SVM-RBF) and K-nearest neighbor (KNN). The proposed method is evaluated using the combination of Physionet Apnea–ECG database and the University College Dublin Sleep Apnea Database. Using a 10-fold cross validation technique, the SVM-RBF classification approach achieves an average sensitivity, specificity, and accuracy of 93.91%, 96.92% and 93.94%, respectively. The results demonstrate that the approach presented is as precise as the best contemporary methods investigated using the same ECG datasets.