An algorithm for sleep apnea detection from single-lead ECG using Hermite basis functions.

This paper introduces a methodology for the detection of sleep apnea based on single-lead electrocardiogram (ECG) of the patient. In the proposed technique, each QRS complex of the ECG signal is approximated using a linear combination of the lower order Hermite basis functions. The coefficients of the Hermite expansion are then used to discriminate the apnea and normal segments along with three features based on R-R time series (mean of R-R intervals, the standard deviation of R-R intervals) and energy in the error of the QRS approximation. To perform classification between the apnea and normal segments, four different types of classifiers (K-nearest neighbor (KNN), multilayer perceptron neural network (MLPNN), support vector machine (SVM), and least-square support vector machine (LS-SVM)) are used in this work. In total, 70 ECG recordings from Apnea-ECG dataset are used in this study and the performance of the proposed algorithm is evaluated based on the minute-by-minute (per-segment) classification, and per-recording (where the entire ECG recording of a subject is discriminated as the apnea or normal one) classification. By considering the events of apnea and hypopnea together, an accuracy of about 84% is achieved on the minute-by-minute basis classification using the LS-SVM classifier with the Gaussian radial basis function (RBF) kernel. On the other hand, an accuracy of about 97.14% is achieved for per-recording classification using the SVM, and LS-SVM classifiers. From the results, it is observed that the proposed methodology provides comparable accuracy with the methods existing in the literature at reduced computational cost due to the lesser number of features selected for the classification.