Znt JBiomed Comput, 25 (1990) 261-272 Elsevier Scientific Publishers Ireland Ltd.
COMPUTER ANALYSIS OF ANTEPARTUM FETAL HEART RATE: 1. BASELINE DETERMINATION
R. MANTEL’, H.P. van GEIJNa, F.J.M. CARON’, J.M. SWARTJES”, E.E. van WOERDEN” and H.W. JONGSMAb #Dept. of Obstetrics and Gynaecology, Academisch Ziekenhuis Vrije Universiteit, De Boelaan I I 17, 1007 MB Amsterdam and bDept. of Obstetrics and Gynaecology, Academisch Ziekenhuis, Nvmegen (The Netherlands)
A consequent and reproducible determination of baseline is an essential prerequisite for objective interpretation of fetal heart rate. A fully automated off-line method of baseline determination has been developed and tested on 50 normal antepartum fetal heart rate recordings of two hours duration. The method is constructed around two functional units, a digital filter and a trim function, which interact in an iterative process. The results were evaluated in comparison with automated baseline determination according to Dawes and coworkers. A panel of 3 experts agreed that in 14 of the 50 recordings (28%), the new developed procedure resulted in a substantially better baseline fit. In the remaining 34 recordings (72Vo), baseline fit from both methods was judged as equivalent. The described procedure of baseline determination provides a solid base for automated detection of accelerations and decelerations in fetal heart rate recordings. It enables the study of the relation between the fetal heart rate pattern and fetal movements. Finally, it provides an objective tool for analysis of variables within the fetal heart rate with the highest predictive value with respect to fetal outcome. Keywords: Antepartum fetal heart rate; Automated analysis; Computer; Baseline; Accelerations; Decelerations
Introduction Antepartum fetal heart rate monitoring has become the predominant method to assess fetal condition in high-risk pregnancies. Interpretation of the fetal heart rate pattern is usually based upon basal fetal heart rate, presence or absence of accelerations and decelerations, and baseline variability. Accelerations and decelerations are defined as deviations from the baseline, while the baseline itself is in fact an imaginary line. Baseline has been stated to represent a form of running average of heart rate in the absence of accelerations and decelerations. During accelerations or decelerations, however, the baseline represents a hypothetical heart rate which might have been observed if the accelerations or decelerations had not occurred [ 11. Because of its hypothetical nature, there is no experimental way to verify whether or not the estimate of baseline is valid. Nevertheless, a consequent and reproducible estimation of baseline is an essential prerequisite for objective description of fetal heart rate recordings. For several years, within our research institute the relation between fetal heart rate and movements has been studied using a computer system . Until recently, 0020-7101/90/$03.50 0 1990 Elsevier Scientific Publishers Ireland Ltd. Published and Printed in Ireland
R. Mantel et al.
analysis was performed semi-automatically, i.e., fetal heart rate initially had to be classified by visual interpretation 121. However, only programmes running fully automatically will be able to describe and evaluate objectively the relation between fetal heart rate and fetal movements and facilitate the study of the effects of certain types of medication on fetal heart rate patterns. Initially, we implemented the procedure described by Dawes et al. . Application of this method to 50 antepartum fetal heart rate recordings with a duration of 2 h each, showed that on several occasions the calculated baseline did not correspond to expectations. This was observed at the beginning of recordings, in the environment of groups of accelerations, especially in case of repeated large swings of heart rate with hardly any segments of basal heart rate in between, and during shifts in basal heart rate. In the current article, a fully automated method for baseline determination that minimizes these problems is described and evaluated. Acquisition and Preprocessing A total of 50 antepartum fetal heart rate recordings of 2 h duration was obtained in uncomplicated near term pregnancies. Fetal heart rate was obtained using either the ECG or wide-range ultrasound. The fetal ECG was obtained from the maternal abdominal wall. A cardiotocograph (Hewlett Packard HP8030) generates flash pulses, corresponding to the R-peaks in the fetal ECG. These pulses were sent to a home-made interface connected to the computer. If it was not possible to attain a fetal ECG of sufficient quality, a wide-range Doppler ultrasound unit with autocorrelation (Hewlett Packard HP8040) was used. This monitor supplies an analog signal which fluctuates with the fetal heart rate. This analog signal is sampled at a rate of 10 times per second and AD-converted. The sample values are subsequently stored on disk. Details of the recording and storage techniques have been described elsewhere . For baseline determination, heart rate signals from a 2-h recording were read from disk and were converted to RR-intervals. Artefact detection according to the algorithm described by Van Geijn et al.  was performed. Evaluation of Baseline Procedure According to Dawes et al. After processing the 50 antepartum fetal heart rate recordings using the procedure of Dawes et al. , there appeared to be 15 recordings in which baseline fit was not optimal regarding one or more of the following aspects: (1) baseline did not fit well at the beginning of the recording (n = 8); (2) baseline level was slightly elevated in the environment of groups of accelerations (n = 7); (3) repeated large swings of heart rate caused a distinct elevation of baseline level (n = 3); (4) baseline did not adequately follow shifts in basal heart rate (n = 6). In the baseline procedure according to Dawes et al. , heart rate signals are averaged over 3.75-s periods after which a digital low-pass filter is processed. The filter programme calculates its starting point from the first 64 samples . In those recordings where heart rate showed a slow and steady declining or inclining
Computer analysis of FHR: Baseline
trend at the beginning, baseline started too low or too high and reached an acceptable level only after a few minutes. A low-pass filter cuts off high frequency components from the signal. Since accelerations and decelerations give rise to very low frequency components which easily pass the filter, they have a substantial influence on baseline level. To diminish the pull that those events assert to the baseline, Dawes et al.  limited the operation of their filter by keeping the output constant if the input signal reached values beyond the range of P f 60 ms. P represents a relative peak in the frequency distribution of the signal, for which either the modal value or a prominent peak near the higher end is chosen. In spite of this precaution, groups of accelerations and large swings of the fetal heart rate appeared to cause an unacceptable elevation of baseline. Narrowing the operating range of the filter does not solve the problem because the filter would no longer adequately cover a physiological drift in basal heart rate over a longer time period. Lowering the cut-off frequency of the filter would make baseline less tractable, resulting in a reduced sensitivity for deviations from basal heart rate. Unfortunately, a cut-off frequency sufficiently low to have any effect would prevent baseline from keeping up with physiological baseline shifts, i.e. long term fluctuations in basal heart rate. To minimize the problems described, one therefore needs another approach. Modified Procedure for Baseline Determination: Filtering and Trimming in an Iterative Process After acquisition and preprocessing, signals were averaged over 2.5-s periods and stored into array A(A,. . . AN). N is the total number of mean RR-intervals thus obtained. Missing values were linearly interpolated between values for which information was available. For programming reasons, values in array A were copied to arrayB(B,. . .BJ. A frequency distribution of mean RR-intervals in array A between 300 and 600 ms was computed using a class width of 1 ms. This distribution was scanned from high to low values in order to locate a relative peak P near the higher end. At least one-eighth of the area of the distribution was scanned before a peak was identified. To be accepted, the peak should exceed the next five classes scanned . The value of P will act as a guide in the baseline procedure as described below. The way in which P was selected, i.e. a prominent peak near the higher end of the frequency distribution, will direct a baseline fit near the lower end of the basal heart rate. The baseline procedure has been constructed around two functional units, a lowpass filter and a trim function, which interact in a five-run iterative process to determine the baseline. The low-pass filter is centered (attenuation 50%) at 0.1 min-’ and is limited in its operation, i.e., the filter only uses those mean RR-intervals that do not deviate more than 60 ms from the previously calculated peak P. To avoid phase shift, the filter processes the signal in either direction. The forward and backward passes convert the signal in array B to a preliminary baseline in array B, while the original signal in array A is preserved. The baseline filter establishes its initial value B, by performing
R. Mantel et al.
a preceding dummy pass in the backward direction. Starting with the value of the relative peak in the frequency distribution, B, is gradually adjusted to obtain the best estimation of the baseline level just prior to the start of the signal. The dummy pass does not influence the contents of array B. Backward dummy pass: B, = P For i running from Nthrough 1: If 1Bi - P(46OthenB, = 0.95 * B, + 0.05 * Bi. Forward pass: For i running from 1 through N If ) Bi - P (