Heart rate variability (HRV) analysis
The electrocardiogram (ECG) reflects the electrical activity of some structures of heart. The depolarization of the atria (the P-wave), the depolarization of the ventricles (the QRS-complex) and the repolarization of the ventricles (the T-wave) are recognized from ECG. Instead the signals of the sinoatrial (SA) node, the atrioventricular (AV) node and the ventricular specialized conduction system (VSCS) are too small to be picked up in routine ECG registrations. Since abnormalities in impulse formation and impulse conduction frequently occur in the "invisible" structures (SA node, AV node, VSCS), ECG interpretation is not always staightforward. Variability in cardiovascular activity, such as RR interval, is widely used as a measure of cardiovascular function.
Heart rate variability is a reliable quantitative marker of autonomic activity. The continuous activity of the autnomic nervous system causes fluctuations into consecutive RR intervals. Heart rate variability can be assessed with time and frequency domain methods. The simplest time domain measure is probably the standard deviation of RR intervals. The frequency content of HRV signal is traditionally divided into three bands: VLF (0 - 0.04 Hz), LF (0.04 - 0.15 Hz) and HF (0.15 - 0.4 Hz). The most popular frequency domain measures are the VLF, LF and HF band powers and the ratio LF/HF.
The procedure for obtaining a reliable power spectral density (PSD) estimate for measured ECG signal is described stepwise as follows.
|1.||A discrete event series RiRi-1 is constructed by an adaptive QRS detector algorithm. The low frequency trend in the series is extracted with appropriate method (e.g. smoothness priors method).|
|2.||In order to recover an evenly sampled signal from the irregularly sampled event series a cubic interpolation should be applied. If the spectrum is however calculated from the unevenly sampled series, assuming it to be regularly sampled, distortion and spurious harmonics are generated on the spectrum.|
|3.||Methods for the calculation of PSD estimate can be classified as nonparametric (e.g. methods based on FFT) and parametric (methods based on autoregressive (AR) time series modelling). With parametric methods the spectrum can be divided to components. These components are due to the roots of the AR polynomial.|