The fetal electrocardiogram (FECG)
The FECG morphology, as illustrated in Fig. 6, is similar to that seen in the adult and contains the P wave, QRS complex and T wave. Fetal cardiac physiology is functionally different from its adult counterpart. In the fetus, the right ventricle plays the dominant role in perfusing the systemic circulation. As a result, the fetal cardiac axis points towards the right in the fetus in contrast to the left-sided deviation in the adults [28]. This difference in orientation results in the FECG appearing morphologically different from an adult ECG [6]. The FECG waveform is processed to provide a familiar ECG visualization to the clinician.
Cardiac time interval (CTI) analysis
A variety of automated computational methods which have been developed for enhanced analysis of the FECG.
Signal detection
The non-invasive nature of the NIFECG relies on signal acquisition from the maternal abdomen. The raw signal consists of the FECG buried within the maternal.
ECG (MECG) signal and environmental noise, such as the uterine muscle activity (UA). Importantly, between the 27th and 32 weeks of pregnancy, the vernix caseosa coats the fetus’ skin and acts as an electrical insulating layer, reducing the efficacy of acquiring the abdominal FECG signal. Signal detection methods have vastly improved over the last 20 years and available technologies at present allow for the detection of the FECG complex from the acquired raw signal [29]. The improvements include and are not limited to new electrode materials (with greater conductivity and skin adhesion), enhanced magnetic shielding of the electronic system, and enhanced electronic designs for noise reduction. The T wave in particular has been known to have lower detection levels due to its weaker signal and distortion by low frequency background noise. As demonstrated by Taylor et al., detection rates of CTIs in the term fetus tend to be more consistent (92%) [25]. Of note are also the presence of physiological conductors and insulators which enhance and attenuate the fetal ECG respectively. Amniotic fluid is an example of a conductor which helps propagate the fetal ECG from the fetal heart to the maternal skin. The vernix caseosa is a sebaceous, protective coating that forms between the 27th to 32nd weeks and persists partially till the 37th week of gestation where it fully dissolves. The vernix caseosa acts as an electrical shield, attenuating the fetal ECG signal. During this period however, a fetal ECG can still be observed non invasively on the mother’s skin as the fetal ECG leaks through the current pathways such as the umbilical cord, oronasal cavity and holes in the vernix caseosa [19, 30].
Signal enhancement
Signal enhancement could be broken down into two distinct steps. Firstly, pre-processing allows the signal to be observed in a suitable frequency range, eliminating artifacts and unwanted features. Secondly, enhancing the FECG and attenuating the maternal ECG is key to accurate FHR and CTIs calculations.
One key function of pre-processing is to narrow the frequency range of the acquired signal. The frequency band would be dependent on the features that need to be seen. If the ST segment needs to be analysed or observed, a lower bound of 0.05 Hz would be needed. If the ST segment is not a concern, for example in ambulatory monitoring of ECG, then 0.5 Hz is a commonly used lower cut off. Devices have commonly used an upper frequency bound of 250 Hz to 1000 Hz which is more than sufficient given the spectrum of the ECG features can be observed below 80 Hz. Therefore, by Nyquist sampling, any sampling frequency above 160 Hz should capture the entire ECG signal with all its features. The upper bound ensures that the sharpest features of the signal can be observed whilst attenuating high frequency noise. The lower bound is meant to cut-off as much baseline drift as possible without compromising low frequency components in the NIFECG signal. Line noise, defined specifically at 50 or 60 Hz, can be cancelled out by a notch filter, which eliminates specific frequencies without altering the rest of the signal. The notch filter frequency has no bearing on the upper or lower bound of the frequency band.
The NIFECG signal, being dominated by the maternal ECG, would require enhancement and processing before it could be utilised for FHR or fetal CTIs [5, 15] . This would mean attenuating the maternal ECG to allow the details of the FECG to be observed. The MECG and FECG occur as independent events and therefore attenuating the maternal signal would not comprise information in the FECG if done accurately.
A variety of techniques exist, such as adaptive filtering, Kalman filtering, Bayesian inference techniques, and de-noising methods [31]. Adaptive and Kalman filtering involve multiple observations of the signal, extraction of the FECG waveform, and suppressing the maternal ECG signal through inference or using the maternal ECG lead as a template. In de-noising methods, the NIFECG signal is decomposed into multiple components and the non FECG components, such as Electromyograms (EMG) and MECG, are set to zero. In the case of true synchronicity in timing of the MECG and FECG, a fetal ECG embedded within a MECG would deform the morphology of the MECG signal. Therefore, methods such using a reference maternal lead containing only MECG to train the algorithms to recognise only the FECG, or mathematically transforming the NIFECG into a space where fetal and MECG can be clearly differentiated would be the solution. If FECG happens to be embedded with the maternal QRS complex, that particular beat may be discarded due to the excessive distortion cause by the large amplitude maternal QRS complex. Ideally, FECG should be extracted from isoelectric portions of the MECG, where there are no MECG features to corrupt the FECG.
Waveform detection
Once the signal enhancement is complete, the FECG signal can be analysed for CTIs given that the PQRST features will be more prominent. However, the FECG signal is still within the noise band of the acquisition devices and has to be enhanced further before the PQRST features can be reliably detected. To achieve further denoising, several beats are averaged which provides a smoothing effect on the signal at the cost of losing minute details on the FECG. The averaging ranges from 10s up to 2 min or up to 1000 beats. After averaging the beats, the PQRST are detected by either identifying the QRS as a high frequency feature and P and T waves as low frequency features. There will be loss of information given beat to beat variability of cardiac events but averaging cardiac cycles makes the assumption that the ECG signal is quasi stationary over short time windows, meaning the features of each cardiac cycle within that short time window remain consistent. Different studies have used different time windows and there is no standard. This assumption must be made and may fail in the event of ectopic beats or paroxysmal arrhythmias. Another method is to differentiate the waves using their slope, amplitude, and width as per the Pan Tompkins algorithm]. The Pan Tompkins algorithm consists of 2 learning phases and one detection phase. The learning phases determines the thresholds and limit values and the detection phase produces a pulse for each QRS complex. [32].
Techniques for fetal ECG enhancement and the effects of noise
The sensitivity of the FECG needs to be considered given the weakness of the signal in comparison to the maternal ECG. By over filtering the signal, several features and CTIs in the FECG may be distorted and become unreliable. When pre-processing, the lower bound of the filter will affect the ST segment. If a filter more than 0.05 Hz is applied, the ST segment’s morphology will be affected and become unreliable for diagnostic purposes. This presents a trade off as filtering in the frequency domain for baseline wander causes distortion of the ST segment. A way around this and subject of potential research would be to identify new transforms where ST segments are preserved whilst eliminating low frequency noise.
When the FECG beats are averaged to remove residual noise, the number of beats used will have an effect. Though the more beats used the cleaner the signal obtained, it also means the P, QRS and T waves will widen and hence provide inaccurate CTI calculations. The CTIs should therefore be viewed in relation to the length of averaging. The MECG and FECG signals are quasi-stationary, which means that beats have similar characteristics over a short period of time whilst the heart reacts to changes in stimuli or physiological conditions. This would mean the widths of the waves as well as relative positions of the P, QRS and T waves with respect to each other would change with varying number of beats used for averaging.
Clinical correlation of CTIs
The focus of this paper will be in relation to the temporal intervals for the FECG, as illustrated on Fig. 6. For the purpose of the following discussion, it must be borne in mind that evidence discussed below is mitigated by the technological limitations applicable to the era in which they were carried out. Furthermore, all data presented below has been derived utilising the FSE. As such, caution should be applied in loosely comparing these findings to modern signal acquisition techniques as well as the NIFECG. Additionally, the number of subjects should also be taken into account when interpreting the findings of individual studies. For instance, Arya et al... (n = 20) demonstrated no correlation between all CTIs and GA.
P wave
This parameter refers to the time interval between the onset and end of the P wave. There has been demonstrable evidence to correlate an increase in P wave duration with cardiac size from 17 weeks of gestation [33] . These were similar to findings in Wacker-Gussman et al. (R = 0.2; P < 0.05) and Chia et al [22, 26]. .
In screening for hypoxia, the utility of P wave duration remains equivocal and unproven. Murray demonstrated P wave duration prior to delivery had a negative correlation with umbilical vein noradrenaline levels (r = − 0.4, p < 0.03) [34]. Conversely though, Jenkins et al. produced results showing no correlation between P wave duration and hypoxic and non-hypoxic fetuses as well [35].
From a technical point of view, there are a number of factors which complicate the process of detecting and interpreting the P waves utilising NIFECG. Firstly, its amplitude is low making the signal detection difficult transabdominally. In addition, the width of the P wave would be affected by the number of beats used in the waveform averaging process. The larger the number of beats, the wider the waveform would become and this would make the calculation of the P wave width unreliable as well.
In this context, the available evidence does not seem to demonstrate a role for utilising the P wave in screening for fetal hypoxia. Taking these technological limitations into account however, further research utilising NIFECG would possibly clarify its role in CTI analysis.
PR and RR interval
This refers to the duration between the onset of the P wave and onset of the R peak which denotes the conduction times from depolarisation of the SA node to conduction through the AV node and Bundle of His. The PR interval tends to be longer in male fetuses in comparison to female fetuses presumably due to weight differences [36] . A temporal relationship between PR interval and GA was also noted by Chia et al., Taylor et al. and Yilmaz et al. in their study [22, 24, 27].
In animal models, studies have demonstrated the lengthening of the PR interval with hypoxia [37, 38]. In the lamb model specifically, PR interval and RR interval lengthening were demonstrated during aortic occlusion in sheep. This was hypothesised to be secondary to a vagal response – since it could be obliterated with the administration of atropine and was not reproducible in premature lambs which do not demonstrate advanced baroreceptor and chemoreceptor responses [39, 40].
In humans however, the PR interval has demonstrated paradoxical results in comparison to the animal model. Murray demonstrated in labouring women that there was no significant change in the mean PR interval through the course of labour. In 59% however, shortening of the PR interval was demonstrated in the last hour of labour but this was within the standard error of measurement (13%). This subgroup though demonstrated a weak correlation (r = 0.2) with umbilical cord gas acidemia [39]. Mohajer et al. also showed a 10% shortening of the PR interval from baseline of compromised fetuses which was however, not statistically significant [41]. In a separate study, he also demonstrated a correlation of the PR interval and umbilical artery pH and lactate (r = − 0.38, p < 0.01 and r = 0.36, p < 0.01) expressed as a ratio index (RI) [42].Physiologically, this could be reflective of the predominant role of catecholamines in the latter stages of labour which influences and delays the conductance of the electrical signal through the AV node.
As such, the role of the PR interval in screening for hypoxia remains unproven and further studies in human would be useful in clarifying its role and the physiological mechanism, if any, in screening for hypoxia.
Several authors have also demonstrated a physiologically inverse correlation between the PR interval and RR interval which becomes positive with evolving acidosis [15, 33, 34, 42, 43].Where the interaction remains continually positive above 20 min, an increased risk of acute fetal compromise has been demonstrated as well [34]. The theoretical basis of this stems from the differential response of the SA node and AV to evolving hypoxia. A vagal cause of this remains unlikely as similar responses can be elicited in mature lambs which have been pre-medicated with atropine [43]. During mild hypoxemia, catecholamine levels become elevated resulting in a concomitant increase in fetal heart rate and a shortening of the PR interval - thereby sustaining the negative relationship between both variables. As the hypoxemia gets progressively worse, the highly oxygen dependent slow sodium channels in the SA node are affected before the fast sodium channels present at the AV node, thereby resulting in a compensatory fall in heart rate and RR interval widening. The catecholamine levels though, continue to rise in line with the evolving hypoxemia thereby continually shortening the PR interval. These synergistic changes would therefore inverse the relationship between both variables to make it positive [39, 44] .
To complicate matters however, Luzietti et al. demonstrated similar inversions in the PR-RR relationship which occurred in all bradycardias below 40 bpm [45]. Westgate et al. further demonstrated similar changes in the relationship during the first 30 min of repetitive umbilical cord compressions in term lamb which however, reverted to negative even in the setting of severe hypoxia. This made them question its discriminative ability and cautioned against potentially misdiagnosing a severely hypoxic fetus as being normal [46].
Based on these findings, two parameters were subsequently trialled in clinical studies in the hope of augmenting existing fetal surveillance parameters. The first was the conduction index (CI) which was a derivative of the Pearson’s correlation between the PR interval and the FHR and calculated every two seconds. Fetal distress was suspected based on a positive relationship establishing for longer than 20 min. The second was termed the ratio index (RI) which was a Z transformed product of the interaction between the FHR and PR interval across the total duration of monitoring undertaken across labour which was computed every 10 s to look for chronic fetal decompensation. Utilising a cut-off of > 4% provided a high specificity of 95.5% and accuracy of 89.4% for cord acidemia [42].
Clinically, Reed et al. were the first to assess the utility of PR interval analysis. In their study the addition of PR interval assessment reduced the utilisation of fetal blood sampling (FBS) from 85.5 to 26.8% which resulted in a 4% reduction of missed acidosis at birth [47] . This was followed by a randomized controlled trial (RCT) carried out by Wijngaarden et al. women were randomised to either routine CTG and labour management or CTG monitoring and PR interval analysis. In the latter, if two of the 3 criteria (abnormal CTG, R > 4% or CI positive for > 20 min) were present, FBS or delivery were to be undertaken at clinician discretion. The study found a significant reduction in the group with PR interval analysis of the number of FBS undertaken, the likelihood of an abnormal FBS, missed cord acidemia at delivery and assisted deliveries for presumed fetal distress [44]. These findings were subsequently followed on by a larger multicentre RCT carried out by Strachan et al. The findings of the study however only demonstrated a non-significant reduction in the group with time interval analysis included [63 (13%) vs 78 (16%)] and no significant difference in identifying cord acidemia or unsuspected cord acidemia [48].
In this context, the available evidence does not seem to demonstrate a significant role for utilising the PR interval in screening for fetal hypoxia. Taking the technological limitations into account however, further research utilising NIFECG would possibly clarify its role in CTI analysis.
From a technical point of view, the widening of the signal due to averaging of the beats will not have an impact on PR measurement since the ratio of PR and RR is considered rather than an absolute measurement. However, if CIs or RIs are being used, the averaging window needs to be carefully considered. For CIs & RIs, since a correlation is calculated every 2 and 10 s respectively, the signal averaging window should not exceed those values.
QRS duration
The QRS duration is a measure of the QRS complex and correlates with the time taken for ventricular depolarisation. The QRS duration is longer in males in contrast to females and is directly correlated with ventricular mass and advancing gestation [2, 8, 49]. These findings were mirrored in Chia et al. and Taylor et al. 2003 [22, 24]. There have been suggestions of its utilisation as a surrogate marker for fetal growth and the diagnosis of fetal growth restriction [49, 50]. Pardi et al. suggested that serial measurements would provide a sensitivity of 81% and specificity of 93% in detecting growth restriction if performed serially [37]. Brambati et al. also investigated its utility in women with haemolytic disease of the newborn and noted its ability to discern between fetuses with worsening prognosticating based on a QRS duration greater than four standard deviations above the mean QRS duration for the gestation [51].
From a clinical point of view, the findings regarding the relevance of the QRS complex are mixed. Some authors have demonstrated QRS widening with cord compression [52, 53] . There has however, been no demonstrable link between perinatal outcomes and the QRS duration as these changes could also be demonstrated in normal labours as well [33, 38, 54].
QT interval/ QTc interval
The QT interval represent the time taken for depolarisation and repolarisation of the ventricles. The QTc corrects the QT interval for extremes of heart rate. In humans, Oudijk et al. noted in their post hoc analysis of 68 fetuses with acidemia at birth the shortening of both the QT and QTc when metabolic acidosis was present and during variable decelerations between the onset and end of labour. They theorised this to be related to a catecholamine effect [53, 55]. Similar findings were noted in the recipient fetus in TTTS - which exhibits myocardial diastolic dysfunction that suggested its utility in identifying deteriorating ventricular performance as well [55]. Paradoxically however, there has also been evidence to suggest that QT interval instead is prolonged with fetal acidosis [15] .
As such, the role of the QT interval in fetal monitoring is yet to be established or resolved.
Areas for further research
The following review highlights several areas to address in terms of future research.
Large scale prospective studies
The present review has identified the necessity for larger scale prospective trials to establish a reliable set of normal CTIs for fetuses across various gestations. This will pave the way for a reliable reference standard to be established in the field. The values presented here in Table 2 would ideally provide a matrix to build future NIFECG studies upon. Ideally, the studies should be grouped in 4 weekly segments (i.e 24–28 weeks, 28–32 weeks) to increase their utility and accuracy. In addition, there would be virtue in exploring technological consistency and validity across these segments as well. Statistical techniques which would aid in interpreting these between group differences would include and are not limited to the intraclass correlation coefficient (ICC) and regression analysis. In comparing the NIFECG, the FSE would be the reference standards for CTI based information as such data cannot be reliably extracted from the CTG. Research direction should also focus on exploring the performance of the CTIs in screening for fetal hypoxia as well. End points of note for hypoxia can be identified from the discussion section of the following review.
Establishing NIFECG databases
Data collected during studies should be combined to form databases to allow investigators in the field to test various algorithms to extract CTIs. Though beyond the scope of this study, Behar et al. provide a reference guide on how to build a standard NIFECG for research purposes which serve as a valuable reference to researchers in the field [31]. This will contribute to conformity and higher quality of data.
Technological consistency
As discussed above and as presented in Table 1, the methods utilised to acquire and process the CTIs are varied in nature and can lead to measurement error bias in the CTI values.
In the context of CTI analysis, consistency between signal processing techniques should be established in order to allow for meta-analysis of data. The averaging of the beats and signal filters in particular need to be considered when performing a meta-analysis to ensure the data is treated within bounds that allow it to be judged as similar. The number of beats or width of the window used for signal averaging is important as a large number of beats or large window will lengthen the CTIs and won’t be representative of the quasi stationary nature of the individual beats. Minimal window sizes of less than 5 s would be preferable due to the high variability associated with fetal heart rate.
Signal filters allow for noise attenuation and enhancement of the signal. However these filters can cause phase delays affecting morphology and temporal alignment between the different leads. Also they eliminate various frequencies which again affect the morphology of the signal, which depending on performance, would affect the signal loss and CTI calculation.
Improving detection methods
Attention should be directed towards improving or overcoming signal attenuation encountered between 28 and 32 weeks in gestation. This can be overcome by adding leads for the pick up of leaked FECG signals. An increased number of leads, greater signal amplification and robust de-noising techniques would aid in improving signal loss during the 28th to 32nd week period. This approach tackles the problem from a signal acquisition, pre-processing and post processing perspective. The greater number of leads would improve the chances of picking up leaked fetal ECG signals which would be directional based on its source, the electronic amplification enhances the signal at point of acquisition and the robust de-noising would enhance the usefulness of each individual lead.
Another issue lies in the lack of gold standard measurements of CTIs used as benchmarking. This would be necessary to establish how accurate are the CTIs, especially in the case of NI-FECG.
Limitations
There were several limitations for the following study. Firstly, given the small amount of data published, there was as limited amount of data for analysis. In particular, the study by Taylor et al. was utilised in patients in labour. Although all foetuses included in the study were normal and no instances of fetal distress/ hypoxia was mentioned in their study, the effect of labour on the CTIs needs to be taken into consideration as it may have affected our results [25]. Also, the wide variation in CTI acquisition techniques and signal processing did not allow for meta-analysis. This would have been useful for examining temporal relationships between the CTIs. In addition, the studies included in the review were at high risk of bias due to study design as well. Nevertheless, the following studies do still demonstrate the benefit and potential in utilising CTIs in fetal diagnostics.