Kullback-Leibler clustering of continuous wavelet transform measures of heart rate variability

Power spectral analysis of beat-to-beat heart rate variability (HRV) has provided a useful means of understanding the interplay between autonomic and cardiovascular functionality. Despite their utility, commonly employed frequency-domain techniques are limited in their prerequisite for stationary signals and their inability to account for temporal changes in the power spectral and/or frequency properties of signals. The purpose of this study is to develop an algorithm that utilizes continuous wavelet transform (CWT) parameters as inputs to a Kohonen self-organizing map (SOM), providing a method of clustering subjects with similar wavelet transform signatures. Continuous interbeat-intervals were recorded (Portapres monitor at 200 Hz) during a perception of affect test in 79 African-American volunteers (ages 21-83), where after a 5-min baseline, participants evaluated emotional expressions in sentences and pictures of faces, followed by a 5-min recovery. Individual HRV biosignals from each session were pre-processed (artifact replacement and signal resampling at 2 Hz) and a CWT was applied (db9 wavelet basis function over 32 scales). Standard deviations of resulting wavelet coefficients at each scale were calculated, normalized, and used as inputs into a SOM with Kullback-Leibler divergence as the dissimilarity measure used for clustering. Differences in subject demographics between two final clusters were assessed via two-independent-groups t-tests or chi-square or Fisher's exact tests of contingency tables. Significant differences were found for age, initial systolic blood pressure, smoking status, and mean s.d. of coefficients in the high frequency band (0.15-0.4 Hz). These findings may have clinical significance and the developed algorithm provides an alternative means of analyzing HRV data originating from populations with complex covariates.