DOI 10.1515/bmt-2011-0130
Biomed Tech 2012; 57(5):383–394
Janusz Jezewski*, Adam Matonia, Tomasz Kupka, Dawid Roj and Robert Czabanski
Determination of fetal heart rate from abdominal signals: evaluation of beat-to-beat accuracy in relation to the direct fetal electrocardiogram Abstract: The main aim of our work was to assess the reliability of indirect abdominal electrocardiography as an alternative to the commonly used Doppler ultrasound monitoring technique. As a reference method, we used direct fetal electrocardiography. Direct and abdominal signals were acquired simultaneously, using dedicated instrumentation. The developed method of maternal signal suppression as well as fetal QRS complexes detection was presented. Recordings were collected during established labors, each consisted of four signals from the maternal abdomen and the reference signal acquired directly from the fetal head. After assessing the performance of the QRS detector, the accuracy of fetal heart rate measurement was evaluated. Additionally, to reduce the influence of inaccurately detected R-waves, some validation rules were proposed. The obtained results revealed that the indirect method is able to provide an accuracy sufficient for a reliable assessment of fetal heart rate variability. However, the method is very sensitive to recording conditions, influencing the quality of signals. Our investigations confirmed that abdominal electrocardiography, even in its current stage of development, offers an accuracy equal to or higher than an ultrasound method, at the same time providing some additional features. Keywords: abdominal signals; fetal electrocardiography; fetal heart rate; fetal monitoring; fetal scalp ECG. *Corresponding author: Janusz Jezewski, Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Phone: +48 322716013, Fax: +48 322715608, E-mail: [email protected] Adam Matonia: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Tomasz Kupka: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Dawid Roj: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland
Robert Czabanski: Institute of Electronics, Silesian University of Technology, 16 Akademicka Street, 44-100 Gliwice, Poland
Introduction The main task of electronic fetal monitoring is to assure optimal conditions allowing the mother to give birth to a healthy and well-developed infant. At present, the only vital function of a fetus that can be recorded effectively is the heart activity signal. Its measurement relies on detection of successive heartbeats and determination of time intervals between them (the so-called RR intervals, hereafter denoted as TRR). However, in clinical practice, the more often used measure is the fetal heart rate [FHR, expressed in beats per minute (bpm)], calculated according to the formula, FHR [bpm] = 60,000/TRR [ms]. One of the most commonly used techniques of FHR measurement is a pulsed Doppler ultrasound method, which detects heartbeats from movements of the fetal heart. The complex structure of the Doppler signal envelope makes precise recognition of R-wave equivalents very difficult [16]. Depending on which episode is recognized as representative for a given heartbeat, different values of cycle duration can be obtained. This influences the accuracy of the TRR period measurement, which is crucial for FHR variability analysis. In case of short-term variability (describing fluctuations of beat-to-beat intervals), even slight distortions of TRR values cause considerable errors in calculation of indices describing this variability [22, 23, 27]. In fact, the exact cardiac cycle can be measured only on the basis of an electrical activity signal – the fetal electrocardiogram (FECG). Recording of the FECG can be accomplished by two methods: the direct method – possible only during labor, where the spiral electrode is directly attached to the fetal head, and the indirect one – where measuring electrodes are placed on the maternal abdomen. The direct method provides a “pure” FECG signal, where the low-frequency interferences can be easily filtered out. However, the most promising from
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
384
J. Jezewski et al.: Determination of the FHR from abdominal signals
the clinical point of view is the indirect method, which has two fundamental advantages over the direct one: it is non-invasive and can be applied during pregnancy. The main problem of its practical implementation is the interfering maternal electrocardiogram (MECG), many times exceeding the signal of interest. A number of different approaches for the suppression of the MECG and the detection of fetal QRS complexes have been presented in the literature [18, 34, 36]; however, thus far, there has been no comprehensive study evaluating the accuracy of FHR measurement in abdominal FECG signal, on a beat-to-beat basis. It should be mentioned that accuracy is a crucial issue in the interpretation of the FECG, as it directly affects the variability assessment. It is of particular importance because the continuous variability of fetal cardiac cycle duration is one of the most important premises indicating appropriate fetal development. In most reports related to detection of the FHR signal in abdominal FECG, the investigations are limited just to an assessment of the success rate of deriving the FHR values from raw abdominal data, or to the assessment of the signal loss ratio [2, 10, 19, 29]. Although a high success rate facilitates the visual assessment, it will be revealed further in this article that it cannot be considered a reliable measure of the method’s performance in case of quantitative analysis. Another common approach in evaluating the quality of abdominal FECG signals is to compare them with the signals obtained by means of the Doppler ultrasound technique. Ibrahimy et al. [13] evaluated 29 min of signal acquired from five patients and noted that 84% of the differences between FHR values obtained using both methods remained inside the ± 5 bpm limit. Similarly, Reinhard et al. [30] obtained 95% of differences within ± 12 bpm. However, Taylor et al. [32] remarked that Doppler ultrasound monitors provide an FHR that is not a true beat-to-beat heart rate but represents an average over neighboring beats. Thus, it is unreliable to consider this method as a reference one, and to evaluate the accuracy of fetal abdominal electrocardiogram the direct FECG should be used. There are only a few reports in which the direct FECG was used as a reference signal. The major shortcoming, limiting the use of a direct FECG technique in labor, is its invasiveness: the needle-like electrode has to be attached directly to the fetal head. This may not be acceptable to obstetricians or the mothers, as fetal scalp injury and perinatal infection from membrane rupture may occur. This kind of study was conducted by Graatsma et al. [9], who analyzed 22 intrapartum recordings of 1 h duration and compared beat-to-beat FHR values as well as indices describing their variability. However, the recording system
had a sampling frequency equal to 300 Hz, which is inadequate for determining the true beat-to-beat variation. Intrapartum recordings of better quality (32 channels at 1 kHz) were acquired by Clifford et al. [4]. However, the main aim of their work was to prove that the FECG signal can be extracted non-invasively without distorting the important clinical parameters, such as the ST-segment. The accuracy was evaluated only by comparison of median FHR values calculated within 10-s epochs. The aim of this study was to evaluate the accuracy of FHR measurement and to estimate its influence on the quantification of beat-to-beat FHR variability. As a result of this investigation, we would like to find out whether abdominal electrocardiography can be considered an alternative to the commonly used Doppler ultrasound monitoring. The next section presents the instrumentation and processing methods being our approach to acquisition and analysis of abdominal FECG signals. The recording module was equipped in an additional channel for acquisition of the direct FECG, which remains a gold standard in FHR measurements. Additionally, an original algorithm for the suppression of MECG is described in this section, allowing us to obtain the original undisturbed locations of fetal QRS complexes. The Results section is organized as follows. First, the performance of our method is assessed using different measures of fetal QRS detection success rate. The obtained results are compared with those reported by others to confirm the reliability of our method. Then, basing on the reference fetal QRS complexes precisely detected in direct FECG signal, we assessed the accuracy of the TRR intervals determination. Analyzing these results, we noticed that the error value of the calculated TRR is strongly affected by a small number of inaccurately detected QRS complexes. To solve this issue and to reduce its influence on the short-term variability indices, we proposed an additional validation procedure. The assessment of accuracy was repeated for the validated signal. Finally, we evaluated the relation between this error value and the error of short-term variability indices proposed by Dalton et al. [6]. All these investigations were conducted with reference to direct FECG, being a gold standard in FHR measurement.
Materials and methods The current work presents a new instrumentation and processing method for recording and analyzing abdominal FECG signals. The system consists of a microcontroller-based signal recorder and a PC [15]. The recorder is
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
J. Jezewski et al.: Determination of the FHR from abdominal signals
equipped with four differential abdominal channels and one for connecting a spiral electrode attached to the fetal head [12]. The electrodes were placed as shown in Figure 1: four around the navel, a reference electrode above the pubic symphysis, and a common mode reference electrode (with active-ground signal) on the left leg. To reduce the skin impedance, the areas under the electrodes were abraded. The number of abdominal channels is a compromise between simplicity of application and the feasibility of signal acquisition, as very often the FECG signal of a good quality is present only in one channel, whereas in the others this signal is not observed [14].
Signal recorder Considering the abdominal FECG, the basic requirement for the signal recorder is a very low level of its own noise (not exceeding 1 μV, measured peak-to-peak with reference to input) as well as high value of common mode
385
rejection ratio (CMRR). These have been obtained with a complete separation of the analog and digital subcircuits [24, 29]. Abdominal and direct FECG are amplified by five instrumentation amplifiers (CMRR = 115 dB). Then, the signals are passed through high-pass filters, whose cutoff frequency can be switched between 0.05 and 1 Hz, depending on the level of low variability interferences. The upper band limit of all signals is established by lowpass filters with a cutoff frequency of 150 Hz. All five signals are consecutively passed through an analog multiplexer to a 16-bit A/D converter and sampled with 1-kHz rate. The RISC microcontroller receives the data from the A/D converter and sends them through an optically isolated RS232 to the PC. The filtering in the analog part is used for elimination of isoline drift as well as for narrowing of the signal bandwidth (anti-aliasing). As the frequency response magnitude of the analog filter is not steep enough, an additional digital filtering is used in a PC to suppress the low-frequency and power line interferences [15].
Figure 1 Instrumentation for recording and analysis of bioelectric signals from the maternal abdomen as well as from an electrode attached to the fetal head. Typical configuration of the abdominal electrodes (A) and a block diagram of the bioelectric signal recorder (B).
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
386
J. Jezewski et al.: Determination of the FHR from abdominal signals
MECG suppression Suppression of the dominant component in abdominal signals – the MECG – is the decisive step in abdominal fetal electrocardiography [3, 5, 33, 35]. The MECG amplitude of about 200 μV is much higher than the FECG amplitude (only about 20 μV). The frequency range of the maternal QRS complex covers the range from 10 to 40 Hz, whereas the fetal QRS complex from 20 to 40 Hz [14]. When analyzing the power density spectrum of the abdominal signal, it can be noted that in the common part, the maternal QRS is still a dominant component, making the suppression of MECG by the simple filtering impossible. In general, the currently proposed method of MECG suppression consists of the following steps: determination of the maternal fiducial points M(i), calculation of the PQRST pattern, and its subtraction from abdominal signal around the fiducial points [26]. To assure a high precision during the subtraction, an error of inaccurate pattern synchronization (resulting from A/D conversion) should be removed. During the subtraction process, the PQRST pattern and the incoming complex are synchronized in time using a correlation function. However, a minimal shift corresponding to sampling period may affect the suppression of the steep slopes of the QRS complex and leave some residual component. If the error is equal to
the sampling period, residual component takes the form of digital first derivative of the QRS complex. This inaccurate pattern of synchronization can be eliminated by subtraction of the derivative of the maternal QRS pattern, scaled by the factor c2, enabling an appropriate amplitude fitting. As the amplitude of maternal QRS complexes change from beat to beat, the pattern (being an average over a number of complexes) has to be scaled each time to properly match the successive complexes. Therefore, the c1 scaling factor was defined, adjusting the amplitude of the QRS pattern and leading to full suppression of the MECG signal. The first step of our algorithm (Figure 2) relies on precise detection of maternal fiducial points. Superimposing of maternal and fetal QRS complexes can cause false determination of M(i), resulting in ineffective MECG suppression as well as unwanted suppression of the FECG. Therefore, abdominal signals after preliminary filtration are used to form an auxiliary signal xSUM, containing MECG exclusively: K
{
xSUM ( n ) = ∑ wSUM , k ⋅ xk ( n ) k =1
}
(1)
where xk(n) is the n-th sample of k-th abdominal signal and wSUM,k is a weighting factor for k-th abdominal signal.
Figure 2 General diagram of the procedure responsible for suppression of the MECG.
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
J. Jezewski et al.: Determination of the FHR from abdominal signals
The wSUM,k factors are calculated on the basis of generalized singular value decomposition [3], to maximize the maternal signal energy and suppress the fetal signal as well as interference components (A). The obtained xSUM signal enables precise determination of fiducial points M(i) (B) and calculation of the c2,i factor (C). The fiducial points enable setting of the maternal QRS pattern, PQRS,SUM, and calculation of its derivative, dPQRS,SUM. Estimation of the c2,i factor is carried out by comparison of the remaining part in the xSUM signal after subtraction of the PQRS,SUM pattern with the derivative dPQRS,SUM. In the next step, the PQRST pattern is established for each signal by averaging of several consecutive maternal cardiac cycles (D):
⎧1 I ⎪ ∑ xk ( n + M ( i )), n ∈ rL ,rP PPQRST , k ( n ) = ⎨ I i=1 ⎪ 0, n ∉ rL ,rP ⎩
(2)
where I is a number of averaged PQRST segments; M(i) is the appropriate fiducial point; rL, rP are values determining the pattern width; and k is a number of particular abdominal signal channel. The most important change in relation to the original algorithm presented in ref. [26] is the introduction of variable width of the maternal PQRST pattern and QRS complex, which are being determined individually for each recording. It is particularly important if the signal is being acquired during labor, when the duration of complexes can be significantly different. Then, a fragment corresponding only to the maternal QRS complex is separated from the pattern and its derivative is calculated (E). For each channel, the derivative is scaled by the weighting factor c2,i and subtracted from the xk signal (F):
xSUB 2 , k ( n + M ( i )) = xk ( n + M ( i )) − c 2 ,i ⋅dPQRS , k ( n ) n ∈ rL ′ ,rP ′
(3)
387
Fetal QRS detection The detection of the fetal QRS complexes is based on determination of fiducial points F(i) corresponding to the R-waves. The time interval between two consecutive fiducial points defines the duration of cardiac cycle TRR. Our online detector of QRS complexes consists of two main blocks (Figure 3). First, the FECG signal covered with noise is fed to the fetal QRS enhancement block, which is based on a digital filter cascade whose frequency response magnitude is adjusted to the spectrum of fetal complexes (Fcl = 12 Hz, Fch = 42 Hz). The non-linear operation consists in modulus function combined with smoothing (moving average of 30 ms width). As a result, a new signal D is obtained, with the same number of samples but of a different shape. A strong, clearly visible peak should point out the R-wave in the original FECG, while outside the QRS complexes the signal values are expected to be close to zero [7, 20, 28]. Then, the output detection signal D is fed to the heartbeat determination block whose task is to detect the peaks and, on the basis of a set of decision rules, to decide whether particular peaks represent the fiducial point or not [1, 11, 21]. The precise location of the i-th QRS complex is obtained as the D maximum value within the segment above the continuously updated detection threshold DT [25]:
F ( i ) = n0 ⇒ D( n0 ) = max D( n ) n1 ( i )
Biomed Tech 2012; 57(5):383–394
Janusz Jezewski*, Adam Matonia, Tomasz Kupka, Dawid Roj and Robert Czabanski
Determination of fetal heart rate from abdominal signals: evaluation of beat-to-beat accuracy in relation to the direct fetal electrocardiogram Abstract: The main aim of our work was to assess the reliability of indirect abdominal electrocardiography as an alternative to the commonly used Doppler ultrasound monitoring technique. As a reference method, we used direct fetal electrocardiography. Direct and abdominal signals were acquired simultaneously, using dedicated instrumentation. The developed method of maternal signal suppression as well as fetal QRS complexes detection was presented. Recordings were collected during established labors, each consisted of four signals from the maternal abdomen and the reference signal acquired directly from the fetal head. After assessing the performance of the QRS detector, the accuracy of fetal heart rate measurement was evaluated. Additionally, to reduce the influence of inaccurately detected R-waves, some validation rules were proposed. The obtained results revealed that the indirect method is able to provide an accuracy sufficient for a reliable assessment of fetal heart rate variability. However, the method is very sensitive to recording conditions, influencing the quality of signals. Our investigations confirmed that abdominal electrocardiography, even in its current stage of development, offers an accuracy equal to or higher than an ultrasound method, at the same time providing some additional features. Keywords: abdominal signals; fetal electrocardiography; fetal heart rate; fetal monitoring; fetal scalp ECG. *Corresponding author: Janusz Jezewski, Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Phone: +48 322716013, Fax: +48 322715608, E-mail: [email protected] Adam Matonia: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Tomasz Kupka: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland Dawid Roj: Department of Biomedical Signal Processing, Institute of Medical Technology and Equipment, 118 Roosevelt Street, 41-800 Zabrze, Poland
Robert Czabanski: Institute of Electronics, Silesian University of Technology, 16 Akademicka Street, 44-100 Gliwice, Poland
Introduction The main task of electronic fetal monitoring is to assure optimal conditions allowing the mother to give birth to a healthy and well-developed infant. At present, the only vital function of a fetus that can be recorded effectively is the heart activity signal. Its measurement relies on detection of successive heartbeats and determination of time intervals between them (the so-called RR intervals, hereafter denoted as TRR). However, in clinical practice, the more often used measure is the fetal heart rate [FHR, expressed in beats per minute (bpm)], calculated according to the formula, FHR [bpm] = 60,000/TRR [ms]. One of the most commonly used techniques of FHR measurement is a pulsed Doppler ultrasound method, which detects heartbeats from movements of the fetal heart. The complex structure of the Doppler signal envelope makes precise recognition of R-wave equivalents very difficult [16]. Depending on which episode is recognized as representative for a given heartbeat, different values of cycle duration can be obtained. This influences the accuracy of the TRR period measurement, which is crucial for FHR variability analysis. In case of short-term variability (describing fluctuations of beat-to-beat intervals), even slight distortions of TRR values cause considerable errors in calculation of indices describing this variability [22, 23, 27]. In fact, the exact cardiac cycle can be measured only on the basis of an electrical activity signal – the fetal electrocardiogram (FECG). Recording of the FECG can be accomplished by two methods: the direct method – possible only during labor, where the spiral electrode is directly attached to the fetal head, and the indirect one – where measuring electrodes are placed on the maternal abdomen. The direct method provides a “pure” FECG signal, where the low-frequency interferences can be easily filtered out. However, the most promising from
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
384
J. Jezewski et al.: Determination of the FHR from abdominal signals
the clinical point of view is the indirect method, which has two fundamental advantages over the direct one: it is non-invasive and can be applied during pregnancy. The main problem of its practical implementation is the interfering maternal electrocardiogram (MECG), many times exceeding the signal of interest. A number of different approaches for the suppression of the MECG and the detection of fetal QRS complexes have been presented in the literature [18, 34, 36]; however, thus far, there has been no comprehensive study evaluating the accuracy of FHR measurement in abdominal FECG signal, on a beat-to-beat basis. It should be mentioned that accuracy is a crucial issue in the interpretation of the FECG, as it directly affects the variability assessment. It is of particular importance because the continuous variability of fetal cardiac cycle duration is one of the most important premises indicating appropriate fetal development. In most reports related to detection of the FHR signal in abdominal FECG, the investigations are limited just to an assessment of the success rate of deriving the FHR values from raw abdominal data, or to the assessment of the signal loss ratio [2, 10, 19, 29]. Although a high success rate facilitates the visual assessment, it will be revealed further in this article that it cannot be considered a reliable measure of the method’s performance in case of quantitative analysis. Another common approach in evaluating the quality of abdominal FECG signals is to compare them with the signals obtained by means of the Doppler ultrasound technique. Ibrahimy et al. [13] evaluated 29 min of signal acquired from five patients and noted that 84% of the differences between FHR values obtained using both methods remained inside the ± 5 bpm limit. Similarly, Reinhard et al. [30] obtained 95% of differences within ± 12 bpm. However, Taylor et al. [32] remarked that Doppler ultrasound monitors provide an FHR that is not a true beat-to-beat heart rate but represents an average over neighboring beats. Thus, it is unreliable to consider this method as a reference one, and to evaluate the accuracy of fetal abdominal electrocardiogram the direct FECG should be used. There are only a few reports in which the direct FECG was used as a reference signal. The major shortcoming, limiting the use of a direct FECG technique in labor, is its invasiveness: the needle-like electrode has to be attached directly to the fetal head. This may not be acceptable to obstetricians or the mothers, as fetal scalp injury and perinatal infection from membrane rupture may occur. This kind of study was conducted by Graatsma et al. [9], who analyzed 22 intrapartum recordings of 1 h duration and compared beat-to-beat FHR values as well as indices describing their variability. However, the recording system
had a sampling frequency equal to 300 Hz, which is inadequate for determining the true beat-to-beat variation. Intrapartum recordings of better quality (32 channels at 1 kHz) were acquired by Clifford et al. [4]. However, the main aim of their work was to prove that the FECG signal can be extracted non-invasively without distorting the important clinical parameters, such as the ST-segment. The accuracy was evaluated only by comparison of median FHR values calculated within 10-s epochs. The aim of this study was to evaluate the accuracy of FHR measurement and to estimate its influence on the quantification of beat-to-beat FHR variability. As a result of this investigation, we would like to find out whether abdominal electrocardiography can be considered an alternative to the commonly used Doppler ultrasound monitoring. The next section presents the instrumentation and processing methods being our approach to acquisition and analysis of abdominal FECG signals. The recording module was equipped in an additional channel for acquisition of the direct FECG, which remains a gold standard in FHR measurements. Additionally, an original algorithm for the suppression of MECG is described in this section, allowing us to obtain the original undisturbed locations of fetal QRS complexes. The Results section is organized as follows. First, the performance of our method is assessed using different measures of fetal QRS detection success rate. The obtained results are compared with those reported by others to confirm the reliability of our method. Then, basing on the reference fetal QRS complexes precisely detected in direct FECG signal, we assessed the accuracy of the TRR intervals determination. Analyzing these results, we noticed that the error value of the calculated TRR is strongly affected by a small number of inaccurately detected QRS complexes. To solve this issue and to reduce its influence on the short-term variability indices, we proposed an additional validation procedure. The assessment of accuracy was repeated for the validated signal. Finally, we evaluated the relation between this error value and the error of short-term variability indices proposed by Dalton et al. [6]. All these investigations were conducted with reference to direct FECG, being a gold standard in FHR measurement.
Materials and methods The current work presents a new instrumentation and processing method for recording and analyzing abdominal FECG signals. The system consists of a microcontroller-based signal recorder and a PC [15]. The recorder is
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
J. Jezewski et al.: Determination of the FHR from abdominal signals
equipped with four differential abdominal channels and one for connecting a spiral electrode attached to the fetal head [12]. The electrodes were placed as shown in Figure 1: four around the navel, a reference electrode above the pubic symphysis, and a common mode reference electrode (with active-ground signal) on the left leg. To reduce the skin impedance, the areas under the electrodes were abraded. The number of abdominal channels is a compromise between simplicity of application and the feasibility of signal acquisition, as very often the FECG signal of a good quality is present only in one channel, whereas in the others this signal is not observed [14].
Signal recorder Considering the abdominal FECG, the basic requirement for the signal recorder is a very low level of its own noise (not exceeding 1 μV, measured peak-to-peak with reference to input) as well as high value of common mode
385
rejection ratio (CMRR). These have been obtained with a complete separation of the analog and digital subcircuits [24, 29]. Abdominal and direct FECG are amplified by five instrumentation amplifiers (CMRR = 115 dB). Then, the signals are passed through high-pass filters, whose cutoff frequency can be switched between 0.05 and 1 Hz, depending on the level of low variability interferences. The upper band limit of all signals is established by lowpass filters with a cutoff frequency of 150 Hz. All five signals are consecutively passed through an analog multiplexer to a 16-bit A/D converter and sampled with 1-kHz rate. The RISC microcontroller receives the data from the A/D converter and sends them through an optically isolated RS232 to the PC. The filtering in the analog part is used for elimination of isoline drift as well as for narrowing of the signal bandwidth (anti-aliasing). As the frequency response magnitude of the analog filter is not steep enough, an additional digital filtering is used in a PC to suppress the low-frequency and power line interferences [15].
Figure 1 Instrumentation for recording and analysis of bioelectric signals from the maternal abdomen as well as from an electrode attached to the fetal head. Typical configuration of the abdominal electrodes (A) and a block diagram of the bioelectric signal recorder (B).
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
386
J. Jezewski et al.: Determination of the FHR from abdominal signals
MECG suppression Suppression of the dominant component in abdominal signals – the MECG – is the decisive step in abdominal fetal electrocardiography [3, 5, 33, 35]. The MECG amplitude of about 200 μV is much higher than the FECG amplitude (only about 20 μV). The frequency range of the maternal QRS complex covers the range from 10 to 40 Hz, whereas the fetal QRS complex from 20 to 40 Hz [14]. When analyzing the power density spectrum of the abdominal signal, it can be noted that in the common part, the maternal QRS is still a dominant component, making the suppression of MECG by the simple filtering impossible. In general, the currently proposed method of MECG suppression consists of the following steps: determination of the maternal fiducial points M(i), calculation of the PQRST pattern, and its subtraction from abdominal signal around the fiducial points [26]. To assure a high precision during the subtraction, an error of inaccurate pattern synchronization (resulting from A/D conversion) should be removed. During the subtraction process, the PQRST pattern and the incoming complex are synchronized in time using a correlation function. However, a minimal shift corresponding to sampling period may affect the suppression of the steep slopes of the QRS complex and leave some residual component. If the error is equal to
the sampling period, residual component takes the form of digital first derivative of the QRS complex. This inaccurate pattern of synchronization can be eliminated by subtraction of the derivative of the maternal QRS pattern, scaled by the factor c2, enabling an appropriate amplitude fitting. As the amplitude of maternal QRS complexes change from beat to beat, the pattern (being an average over a number of complexes) has to be scaled each time to properly match the successive complexes. Therefore, the c1 scaling factor was defined, adjusting the amplitude of the QRS pattern and leading to full suppression of the MECG signal. The first step of our algorithm (Figure 2) relies on precise detection of maternal fiducial points. Superimposing of maternal and fetal QRS complexes can cause false determination of M(i), resulting in ineffective MECG suppression as well as unwanted suppression of the FECG. Therefore, abdominal signals after preliminary filtration are used to form an auxiliary signal xSUM, containing MECG exclusively: K
{
xSUM ( n ) = ∑ wSUM , k ⋅ xk ( n ) k =1
}
(1)
where xk(n) is the n-th sample of k-th abdominal signal and wSUM,k is a weighting factor for k-th abdominal signal.
Figure 2 General diagram of the procedure responsible for suppression of the MECG.
Brought to you by | Universitaet Giessen Authenticated Download Date | 6/11/15 9:53 AM
J. Jezewski et al.: Determination of the FHR from abdominal signals
The wSUM,k factors are calculated on the basis of generalized singular value decomposition [3], to maximize the maternal signal energy and suppress the fetal signal as well as interference components (A). The obtained xSUM signal enables precise determination of fiducial points M(i) (B) and calculation of the c2,i factor (C). The fiducial points enable setting of the maternal QRS pattern, PQRS,SUM, and calculation of its derivative, dPQRS,SUM. Estimation of the c2,i factor is carried out by comparison of the remaining part in the xSUM signal after subtraction of the PQRS,SUM pattern with the derivative dPQRS,SUM. In the next step, the PQRST pattern is established for each signal by averaging of several consecutive maternal cardiac cycles (D):
⎧1 I ⎪ ∑ xk ( n + M ( i )), n ∈ rL ,rP PPQRST , k ( n ) = ⎨ I i=1 ⎪ 0, n ∉ rL ,rP ⎩
(2)
where I is a number of averaged PQRST segments; M(i) is the appropriate fiducial point; rL, rP are values determining the pattern width; and k is a number of particular abdominal signal channel. The most important change in relation to the original algorithm presented in ref. [26] is the introduction of variable width of the maternal PQRST pattern and QRS complex, which are being determined individually for each recording. It is particularly important if the signal is being acquired during labor, when the duration of complexes can be significantly different. Then, a fragment corresponding only to the maternal QRS complex is separated from the pattern and its derivative is calculated (E). For each channel, the derivative is scaled by the weighting factor c2,i and subtracted from the xk signal (F):
xSUB 2 , k ( n + M ( i )) = xk ( n + M ( i )) − c 2 ,i ⋅dPQRS , k ( n ) n ∈ rL ′ ,rP ′
(3)
387
Fetal QRS detection The detection of the fetal QRS complexes is based on determination of fiducial points F(i) corresponding to the R-waves. The time interval between two consecutive fiducial points defines the duration of cardiac cycle TRR. Our online detector of QRS complexes consists of two main blocks (Figure 3). First, the FECG signal covered with noise is fed to the fetal QRS enhancement block, which is based on a digital filter cascade whose frequency response magnitude is adjusted to the spectrum of fetal complexes (Fcl = 12 Hz, Fch = 42 Hz). The non-linear operation consists in modulus function combined with smoothing (moving average of 30 ms width). As a result, a new signal D is obtained, with the same number of samples but of a different shape. A strong, clearly visible peak should point out the R-wave in the original FECG, while outside the QRS complexes the signal values are expected to be close to zero [7, 20, 28]. Then, the output detection signal D is fed to the heartbeat determination block whose task is to detect the peaks and, on the basis of a set of decision rules, to decide whether particular peaks represent the fiducial point or not [1, 11, 21]. The precise location of the i-th QRS complex is obtained as the D maximum value within the segment above the continuously updated detection threshold DT [25]:
F ( i ) = n0 ⇒ D( n0 ) = max D( n ) n1 ( i )
