Suguru Sugimoto

Lowpass filtering method for heart rate variability analysis

Various methods have been proposed for the spectral analysis of heart rate variability (HRV). Such methods always require constant period data in time series. However, since the R-R interval of HRV is, in general, irregular, somehow the experimental data must be converted to constant period data. Assuming an integral pulse frequency modulation (IPFM) model, constant period data can be obtained by passing the heartbeat impulse train through a digital filter using the alias-free method proposed by French and Holden. However, the forementioned method has two problems; that is: (1) the frequency characteristics of the lowpass filter are assumed to be ideal, hence the impulse response function of the filter spreads infinitely with respect to time; and (2) only pulses inside the analyzing section are considered while those outside the analyzing section in the convolution integral are ignored. To solve those problems, the method of French and Holden has been modified in the following three points: (1) such a finite impulse response digital filter was used that the sidelobe of the filter decays rapidly in its impulse response function; (2) the analyzing section as well as its impulse response time were lengthened, thus the pulses outside the analyzing section were considered; and (3) an appropriate window was used. With these improvements, the truncation error and the computation time could be reduced. Furthermore, as a result of simulation, a much higher accuracy in the demodulation of the heartbeat signal was obtained.

Time-Frequency Analysis of the Blood Flow Doppler Ultrasound Signal

Power spectral analysis is extensively used to interpret ultrasound data. However, the technique is useful only when the data can be treated as stationary. Ultrasound data are mostly nonstationary. Thus, a short time Fourier transform (STFT or spectrogram) is widely used to analyze spectral components which change with time. However, the STFT has a low accuracy in both time and frequency domains. Currently, Cohens class time-frequency (TF) analysis is popular for analyzing nonstationary signals. The authors recently proposed a new kernel (named a figure eight kernel). In order to apply the TF analysis with the new kernel to a blood flow signal, experimental data were obtained from the carotid artery by an ultrasound Doppler monitor (Toitsu, Japan). To analyze the data, three kernels were used: (1) a Wigner kernel, (2) a Choi-Williams kernel, and (3) a figure eight kernel. Using our new figure eight kernel, the demodulation accuracy was improved and blood flow components were observed.

Measurement characteristics of the ultrasound heart rate monitor

It was reported that the measurement error of the fetal heart rate variability (FHRV), which was obtained by a ultrasound heart rate monitor with the Doppler signal, was large even if the auto-correlation technique was used. Nevertheless, fetal heart rate monitoring by the ultrasound heart rate monitor is necessary to determine the status of the fetus because an invasive test cannot be used daily. In order to make sure the quality of the FHRV obtained from the Doppler data, we measured the fetal ECG directly from the fetal sculpture at the same time as the Doppler data. The FHRV differences of the Doppler data from the direct ECG data were found to be concentrated at 0 bpm (beats per minute), around which the pattern of distribution is practically symmetrical. Furthermore, the spectral density of the FHRV differences showed the white spectrum without dominant peaks.<<ETX>>

Estimation of measurement characteristics of ultrasound fetal heart rate monitor

Ultrasound fetal heart rate monitoring is very useful to determine the status of the fetus because it is noninvasive. In order to ensure the accuracy of the fetal heart rate (FHR) obtained from the ultrasound Doppler data, we measure the fetal electrocardiogram (ECG) directly and obtain the Doppler data simultaneously. The FHR differences of the Doppler data from the direct ECG data are concentrated at 0 bpm (beats per minute), and are practically symmetrical. The distribution is found to be very close to the Students t distribution by the test of goodness of fit with the chi-square test. The spectral density of the FHR differences shows the white noise spectrum without any dominant peaks. Furthermore, the f-n (n>1) fluctuation is observed both with the ultrasound Doppler FHR and with the direct ECG FHR. Thus, it is confirmed that the FHR observation and observation of the f-n (n>1) fluctuation using the ultrasound Doppler FHR are as useful as the direct ECG.

Frequency component of fetal heart rate with integral function method

The authors have been studying the control mechanism of the fetal heart rate using a spectral analysis method which has high accuracy and a short calculation time. They have shown by simulation that frequency components of heart rate variability can be obtained more accurately even if the heart rate data contained certain errors in measurement. They report here the result obtained by applying a similar spectral analysis method to human cardiotocographic data using both the ultrasonic Doppler and the direct scalp electrode method and the integral function method for the analysis. The respiratory component is obtained by spectral analysis.<<ETX>>

Time-frequency analysis of Doppler sinusoidal shift signal with new figure eight kernel

In order to apply the Cohens class Time-Frequency (TF) analysis using a novel kernel to the blood flow signal analysis, we investigated a higher resolution kernel for the ultrasound Doppler sinusoidal shift signal. We used three kinds of kernel, the Wigner distribution, the Choi-Williams distribution (CWD), and our new figure eight kernel (FED). The results show that the demodulation accuracy with FED is higher and the negative components are smaller than the CWD.

Highly accurate demodulation method of an IPFM model with an absolute refractory period

An integral pulse frequency modulation (IPFM) model is a pulse generation mechanism model for the nervous systems and is one of the models which connects heart rate variability to autonomic nervous system activity. There is a refractory period just after a heart beat impulse occurs, in which no heart beat impulses occur at any rate. The IPFM model does not take the refractory period into account. It would be good to consider the refractory periods in order to make the IPFM model realistic. In this paper, we are examining the effects of the absolute refractory period on the spectral distortion properties and the demodulation accuracy. Spurious components, (which are caused by mutual interference among the frequency components), decreased as the absolute refractory period increased, while the side-band distortion around the input frequency components increased. The direct FFT method impairs demodulation accuracy as the absolute refractory period increases. Even the integral function (IF) method without taking the absolute refractory period into account can reduce the distortion that is peculiar to the absolute refractory period. Moreover, the IF method which took the absolute refractory period into account has higher demodulation accuracy in spite of the absolute refractory period.

Confirmation of the Distribution of Measured Values from an Ultrasound Fetal Heart Rate Monitor

To estimate the measurement error in an ultrasound fetal heart rate (FHR) monitor, we compared the differences in the distribution of the FHR measured with an ultrasound Doppler to the FHR measured with a direct fetal electrocardiogram. We confirm that (1) the distribution is very similar to Students t distribution, and (2) the number of degrees of freedom in Students t distribution has a significant impact on the evaluation of fit.

Effect of Measurement Error of Cardiotocograph to Spectral Analysis

It is reported that the measurement error of the fetal heart rate variability, which is obtained by a cardiotocograph with a Doppler ultrasonics signal, is large even if the auto-correlation technique is used. Simulations are performed to estimate the effect of the measurement error on the spectral analysis. The following two respects are confirmed: (1) when the error is distributed uniformly from -15 msec to +15 msec, the noise level of the spectrum is -12 dB, and (2) the noise level is not greatly different whether the distribution of the measurement error is uniform or Gaussian.

Time-frequency analysis of a complicated Doppler signal with two figure eight kernels

A Cohens class time-frequency analysis has high resolution on the TF plane both in time and in frequency. A complicated (nonstationary and/or nonlinear) Doppler signal was obtained from synthesized data. In order to analyze this signal, two kinds of figure eight kernels were used:(1) a figure eight kernel, and (2) a 2nd figure eight kernel. As well known conventional kernels, we used a Wigner kernel, and a Choi-Williams kernel. Using our figure eight kernels, the original signal was demodulated with high resolution. Furthermore, a 2nd figure eight kernel improved the demodulation accuracy.