Medical Blog

The frequency analysis of neonatal blood pressure signals differs in many respects from the analysis of signals from adults. In neonates:

• the frequencies of heartrate and respiration are normally much higher,
• the heartrate variability turns out to be less pronounced,
• the number of artefacts in the signals is higher than in adults.

Many studies have been done on heartrate variability in adults and neonates, some on blood pressure variability, but only a few on blood pressure variability in neonates.
The analysis methods used, mainly Fourier transform or a modification of it, and autoregressive model estimates, lead to a variety of representations of the results. All methods have their own characteristics, constraints, and assumptions. In time domain analysis many different indices have been proposed, often only to be used in specific circumstances. The frequency domain methods seem more general; the autoregressive methods require certain model assumptions, like the order of the system. The main problems of the Fourier methods will be discussed in the next sections.

Aliasing

In our analysis the problem of aliasing occurs if the momentary respiratory frequency is higher than half the heartrate. In neonates this situation will occur regularly. The ratio of the respiratory frequency and heartrate in neonates is often higher than in adults. All Fourier techniques are hindered by a relatively high respiratory frequency.
The Lomb method, which is able to handle non-equidistantly sampled signals, shows better results if the major part of the momentary sampling frequencies is larger than half the heartrate.
We found a more fundamental alternative in using the blood pressure filtering (BPF) technique. Using the full blood pressure signal, aliasing will occur only above 32 Hz (half the sampling frequency), which is far beyond our region of interest. The BPF technique is not a final solution. It introduces a problem that is related to the complexity of a physiological signal like the blood pressure signal. The blood pressure
signal consists of pulses that result from different states of heart contraction under strongly changing internal physiological states. These physiological states vary with the heart frequency, and, as a result, they modulate the other variations in the blood pressure signal. Sum and difference frequencies will be present; the difference frequencies being at the same frequency as the aliased frequencies from above half the heartrate when a Fourier technique is used. Using the BPF method, however, they can be explained by a real physical variation. Using one of the other methods, one should always check separately on the respiratory frequency to be attentive to possible spectral aliasing.

Spectral leakage

Spectral leakage implies the leakage of spectral power to neighbouring frequency bins, due to the (generally) non periodicity of the signal in the time domain. In most cases, the application of one of the standard lag-windows gives acceptable results. The HF frequency components we are interested in contain variations, which lead to abroadening of the concerned frequency components. The effect of leakage to the neighbouring bins in that case only marginally affects the result. The LF and VLF components, however, sometimes extend over only a few bins. In that situation one should be aware that the accuracy of the spectral components may decrease considerably.