Automated characterization of sudden cardiac death by using empirical mode decomposition and fuzzy entropy of ECG signals

Abstract

Early prediction of Sudden Cardiac Deaths (SCD), as a promising methodology to save lives threatened by SCD, has garnered significant attention from various countries. Various approaches has been proposed to forecast the occurrence of SCD, these approaches are limited either in terms of prediction time resolution or the length of early prediction time. In this work, we propose a novel methodology that combines Empirical Mode Decomposition (EMD) and Fuzzy Entropy (FuEn) using only one ECG beat with 651 data samples for the early anticipation of SCD. The experiment extracted a total of 652 features, consisting of 651 first-level intrinsic mode function (IMF) coefficients and 1 FuEn extracted from the first-level IMF coefficients. Subsequently, the Locality Preserving Projection (LPP) algorithm is applied to these features to reduce the number of features, the LPP features were ranked by various ways. Finally, the ranked features were fed to classifiers for discrimination between normal and prone-to-SCD subjects. Unlike other works where 1 min or 2 min signal analyzed as the prediction resolution is employed and the length of early prediction time is no more than 20 min, the proposed methodology analyzes each ECG beat for the SCD identification and is capable of predicting SCD up to 30 min earlier with the accuracy, specificity, and sensitivity of 98.5%, 98.4%, and 98.6% respectively. Therefore, the proposed methodology can be applied in portable medical sensors to real-time monitor conditions of person at risk of SCD in hospital settings or at home.