Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 184

FIGURE 4.33: Wavelet analysis of HRV of patient with myocardial infarction.
From top: signal, time-frequency energy distribution obtained by wavelet trans-
form, low frequency power (LFP), high frequency power (HFP), ratio LFP/HFP.
From [Toledo et al., 2003].

4.2.2.4 Non-linear methods of HRV analysis

Given the complexity of the mechanisms regulating heart rate, it is reasonable to
assume that applying HRV analysis based on methods of non-linear dynamics will
yield valuable information. The study concerning the properties of heart rate variabil-
ity based on 24 hour recordings of the 70 subjects was conducted by [Urbanowicz
et al., 2007]. It was reported that variability due to the linearly correlated processes
was dominant (in normal group 85%), but with the development of the disease and
risk of cardiac arrest the non-linear component increased. The contribution of the
random noise variance was found at 5-15% of the overall signal variance. The ex-
ception was the case of atrial fibrillation where the contribution of random noise
achieved 60%. One of the first applications of the non-linear methods was compu-
tation of the correlation dimension of the HRV [Babloyantz and Destexhe, 1988].
The most frequently used parameters that have been applied to measure non-linear
properties of HRV include the correlation dimension, Lyapunov exponents, and Kol-
mogorov entropy. In spite of multiple attempts conducted with the use of the above
methods to demonstrate the prevalence of chaos in heart rate series, more rigorous
testing based primarily on surrogate data techniques did not confirm this hypothesis
and revealed the weakness of classical non-linear methods connected with lack of
sensitivity, specificity, and robustness to noise [Poon, 2000].
More recently scientists have turned to the methods which are less biased by noise
and restricted length of the available data, namely: empirical mode decomposition,