Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 182

recordings the problem of stationarity arises. If the modulations of frequencies are
not stable, the interpretation of the results of frequency analysis is less well defined.
In particular, physiological mechanisms of heart period modulations responsible for
LF and HF power components cannot be considered stationary during the 24-hour
period, thus they should be estimated on the basis of shorter time epochs. To at-
tribute individual spectral components to well-defined physiological mechanisms,
such mechanisms modulating the heart rate should not change during the recording.
Transient physiological phenomena should be analyzed by time-frequency methods.
To check the stability of the signal in terms of certain spectral components, tradi-
tional statistical tests may be used.
Both, non-parametric and parametric methods are used for frequency analysis of
HRV. The advantages of the non-parametric methods are: the simplicity of the algo-
rithm used (fast Fourier transform) and high processing speed, while the advantages
of parametric methods are: smoother spectral components that can be distinguished
independent of pre-selected frequency bands, straightforward post-processing of the
spectrum with an automatic calculation of low- and high-frequency power compo-
nents (easy identification of the central frequency of each component), and an accu-
rate estimation of PSD even on a small number of samples which is important for
the quasi-stationary signal. A disadvantage of parametric methods may be the need
to determine the model order, but this can be achieved by means of known criteria
(Sect. 2.3.2.2.2).
The HRV signal with a period of artery occlusion is shown in Figure 4.32 a).
The HRV spectral variability is illustrated in Figure 4.32 b). The running power
spectra were obtained by means of the adaptive AR model with recursively fitted
parameters [Mainardi et al., 1995]. In the above work the quantification of changes
in HRV was performed by means of tracking the poles of transfer function (Sect.
2.3.2.2.4).
For determination of ECG derived respiratory information spectral analysis is usu-
ally applied. The respiratory activity is estimated as the HF component of HRV sig-
nal. The respiratory rhythm may be determined as the central frequency of the HF
peak of the power spectrum or in case of AR model directly as a frequency deter-
mined by means of FAD (Sect. 2.3.2.2.4 ). The time-varying AR model was used for
analysis of coupling of respiratory and cardiac processes [Meste et al., 2002]. Res-
piratory activity may also be found on the basis of beat morphology, namely from
amplitude fluctuation of main complexes of ECG [Bailon et al., 2006], or possibly
both methods can be combined [Leanderson et al., 2003].
The time-varying features of HRV may be studied by means of wavelet anal-
ysis. Both, continuous, e.g., [Toledo et al., 2003] and discrete wavelet analysis,
e.g., [Roche et al., 2003] were applied in HRV analysis. Wavelet analysis of HRV
of a patient with myocardial infarction is shown in Figure 4.33. Alternation of low
(LF) and high (HF) frequency power is visible after the time of reperfusion. The
advantage of discrete wavelet transform is the parametric description of the signal,
which is useful for further statistical analysis and classification.