Wavelet-Based Multiscale Sample Entropy and Chaotic Features for Congestive Heart Failure Recognition Using Heart Rate Variability
This study explores the discrimination power of a multiscale analysis method based on the discrete wavelet transform (DWT) in characterizing nonlinear features for congestive heart failure (CHF) recognition. Two DWT paradigms, namely standard DWT and DWT with reconstruction (RDWT), were employed to characterize two categories of nonlinear features, namely sample entropy (SE) and chaotic features, for CHF recognition based on heart rate variability (HRV). The performance of the wavelet-based analysis methods was compared to that of a traditional coarse grained average (CGA) method. The support vector machine was used as a classifier and the capability of the features was evaluated using the leave-one-out cross-validation method. The results show that when using solely SE features, all three multiscale analysis methods (CGA, DWT, and RDWT) with five dyadic scales outperform traditional CGA with twenty consecutive scales in characterizing HRV for CHF recognition. When using chaotic features calculated from the five dyadic scales, RDWT outperformed DWT and CGA with sensitivity, specificity, and accuracy rates of 95.45, 97.22, and 96.55 %, respectively. This performance was even superior to that obtained using both SE and chaotic features. The proposed multiscale analysis method using 5-scale RDWT and chaotic features outperforms three well-known CHF classifiers reported in the literature.