Detection of Congestive Heart Failure by Autocorrelation Analysis of Heart Rate Variability

Abstract

This paper introduces the autocorrelation analysis of heart rate variability (HRV) signals to diagnose cardiac diseases and presents a methodology for the automated detection of congestive heart failure (CHF). The autocorrelation is defined as the correlation (a mutual relationship between two or more processes) of a signal with a delayed version of that signal. The proposed methodology has been tested on signals acquired from physionet open access database for clinical relevance. The HRV signals of normal and CHF subjects have been divided into segments of 1000 samples each. Next, the Pearson’s correlation, Rank Correlation (Kendall’s rank correlation and Spearman’s rank correlation), and Partial correlation coefficients have been computed with a delay of τ = [1, 2, … , 10] samples. The clinically significant features obtained by performing one way ANOVA (Analysis of Variance) are fed to decision tree (DT), and neural network (NN) classifiers. The classification results show an accuracy of 87.6% with sensitivity and specificity of 84.9% and 90.3%, respectively. The experiment results reveal that the autocorrelation in HRV signals of normal and CHF subjects is significantly different and suggests a possibility of the diagnosis of CHF from these features.