Uncovering non-linear structure in human ECG recordings

We employ surrogate data techniques and a new correlation dimension estimation algorithm, the Gaussian kernel algorithm, to uncover non-linearity in human electrocardiogram recordings during normal (sinus) rhythm, ventricular tachycardia (VT) and ventricular fibrillation (VF). We conclude that all three rhythms are not linear (i.e. distinct from a monotonic non-linear transformation of linearly filtered noise) and have significant correlations over a period greater than the inter-beat interval. Furthermore, we observe that sinus rhythm and VT exhibit a correlation dimension of approximately 2.3 and 2.4, respectively. The correlation dimension of VF exceeds 3.2. The entropy of sinus rhythm, VT and VF is approximately 0.69, 0.55, and 0.67 nats/s, respectively. These results indicate that techniques from non-linear dynamical systems theory should help us understand the mechanism underlying ventricular arrhythmia, and that these rhythms are likely to be a combination of low dimensional chaos and noise.