Medical Blog

Spectral equidistancy

A signal built of beat-by-beat parameters (a set of consecutive beat-by-beat values) is inevitably non-equidistant in time. In most of the neonatal signals, the consequences will not be dramatic. The 30% distortion of spectral components mentioned by TenVoorde [TenVoorde, 1994] does not apply to neonates because of their small respiratory sinus arrythmia (see section 5.4), and thus small frequency modulation of the sampling intervals of the beat-to-beat values. If the heartrate variability is larger, the error increases. We used neonatal data to evaluate the impact of the heartrate variability on the spectral components, using several different Fourier algorithms. Our conclusion is that the algorithms that specifically take into account the heart interval (EFT, RFT) give (slightly) different results, compared to the standard DFT/FFT algorithm. The standard DFT/FFT was applied such that the non-equidistancy was ignored. Many institutions follow that method; the boxcar integration we apply in the standard procedure introduces an extra filtering and resampling. The extra filtering reduces the aliasing of higher frequencies and enables us to resample equidistantly. The neonatal data we examined show only small differences with respect to the particular Fourier method chosen. Important is that the respiratory sinus arrhythmia, the main cause of the non-equidistancy of the beat-to-beat values, is much smaller in neonates than in adults.

Stationarity

Stationarity is an important item. If one wants to characterise certain spectral components, resulting from physiological processes in the body. If the underlying physiological processes change during the registration, qualitative characterisation will sometimes be impossible, and quantitative characterisation will only be possible in proportion. Given the fact that all physiological processes in the body change in time, it is desired to limit the registration time to a period in which the body processes hardly change. A trade-off has to be made is between the accuracy of the measurement and the stability of the underlying processes. Physiological changes that we want to exclude from the measurements are, for instance,
−changes in the heartrate base level, e.g., caused by labour or mental stress
−changes in behavioural state or sleep state
−changes in the chemical state of the blood (e.g., pCO2 level changes that drive many central processes)
Tools that check the data on stability should be developed. The clinician could then be informed automatically if the observed variations were the result of a changing process during the measurement. We suggest to use correlations between the signals to detect artefacts, but did not work out procedures by now.

Beat-to-beat values versus the full sampled signal

Usually, beat-by-beat values are used in spectral analysis of the blood pressure variations. Originally, the main reason for this approach was that no calculation power was available to handle the frequency analysis of a full sampled signal, consisting of approximately 30 to 100 times as much data as the beat-by-beat values in a certain time interval. We showed that nowadays the limited calculation power no longer impedes the analysis of the full sampled signal. In that case, problems concerning aliasing and non-equidistancy do not exist any more. The graphs in chapter 5 show such results, and a comparison with the results of the analysis of a beat-to-beat signal. Theoretically, a significant improvement of the results should be obtained, if neither aliasing nor non-equidistant sampling would occur. Part of the improvement, however, is spoiled by the modulations expressed in the full sampled signal at sum and difference frequencies of the heartrate and respiratory rate. It will strongly depend on the application and signal characteristics which method will satisfy most.