Monitoring fetal heart rate (FHR) during labor is a common clinical routine worldwide, aiming to asses fetal well-being and ensure safe delivery. The main objective is to decide on timely operative delivery or uterine relaxation to prevent brain injury and adverse outcomes (1). In clinical practice, fetal well-being is assessed by obstetricians principally by visual inspection of cardiotocograms (CTG, bivariate time series of beat-per-minute FHR and uterine activity). The interpretation is guided by a set of rules combining a collection of features, aiming to probe various aspects of the CTG, such as baseline FHR, FHR variability and deceleration shape and timing as well as the relation of the various FHR features to the patterns of uterine activity. One such set of features and rules was defined by the International Federation of Gynecology and Obstetrics (2, 3). Applying such procedure has however been documented as yielding significant inter-and intra-observer variability (4), one of many causes of the failure of the present FHR monitoring to predict fetal brain injury (5–7).

These short-comings in FHR monitoring during labor triggered significant efforts to develop computerized and automated assessment of FHR patterns intrapartum. Beyond the direct computation of the FIGO features themselves [cf., e.g., (8–10)], from digitized CTG usually sampled at 4 Hz in clinical practice, a large variety of features stemming from advanced signal processing and information theory tools has been computed for FHR assessment. These advanced features, however, have not reached performance benchmarks to lead to a consensus in the research and medical communities. These observations leave open a significant number of issues ranging from the choice of relevant FHR features and the construction of decision rules for such features to the assessment of the relationships between FHR time series and fetal well-being. Interested readers are referred to Georgieva et al. (11) (and references therein) for a recent (lack of) consensus overview and the interdisciplinary discussions.

Besides the need for large labeled databases to make machine learning on FHR data effective (12), a recurrent issue is associated with the ground truth being based on pH from the immediate post-birth umbilical cord pH measurements. However, it has been documented that fetal brain injury poorly correlates with measures of acidemia at birth such as pH (7, 11, 13). First, pH is only available after delivery hence when FHR is no longer available. Second, brain compromise due to hypoxia-ischemia can ensue when the fetal cerebral blood flow is persistently reduced, e.g., due to precipitous drop in cerebral perfusion pressure resulting from cardiovascular decompensation (CVD) (7, 14, 15).

In a recent series of experiments, to better assess the relations between FHR, systemic arterial blood pressure (ABP) and fetal health state (including the impact of chronic hypoxia) sheep fetuses were surgically instrumented and subjected to an umbilical cord occlusion (UCO) protocol in Frasch et al. (16), in a well-established animal model of human labor. CVD onset was observed at individually variable times, regardless the presence of chronic hypoxia (7, 16–19). This animal experimental model generated the dataset used in the present study.

Consequently, based on this dataset, the goal of the present work is to assess whether FHR monitoring permits detection of the individual onset of CVD accounting for the presence of chronic hypoxia prior to the onset of UCOs in some fetuses. More particularly, we aimed to assess the sensitivity of FHR temporal dynamics, probed by four scale-dependent features, to CVD, metabolites and pH measurements.

We propose four features which all have the temporal scale of the signal as a parameter. We compute these four quantities from the whole FHR signal to probe its dynamics along the complete experiment. The first quantity measures the average variation of the FHR over the prescribed time scale. The second one measures the FHR variability over the time scale as the standard deviation. The third one is the ratio of the first two and provides a normalized version of the average variation. The fourth one is very similar to Approximate Entropy or Sample Entropy and provides a measure of the information content of the FHR signal at the given time scale. These four quantities can be computed with any signal and give robust results even with the clinically relevant low sampling rate of 4 Hz. This feature choice is also designed to allow computation within short-time time windows, thus permitting to achieve a sliding-window, time-dependent analysis of FHR, which may eventually be exploited to perform real-time FHR monitoring on noisy data.

We show the importance of coarse time scales (2.5–8 s) and construct an individual self-referencing "distance to healthy state" metric based on combination of the four features. We then demonstrate the use of the novel composite distance metric to predict individual CVD from FHR time series alone.

Fetal sheep model of labor and surgical preparation. The anesthetic and surgical procedures, postoperative care of the animals and the UCO model of labor have been previously described Frasch et al. (16). Briefly, 14 near-term ovine fetuses (123 ± 2 days gestational age (GA), term = 145 days) of the mixed breed were surgically instrumented. Animal care followed the guidelines of the Canadian Council on Animal Care and was approved by the University of Western Ontario Council on Animal Care.

Polyvinyl catheters were placed in the right and left brachiocephalic arteries, the cephalic vein, and the amniotic cavity. The fetal arterial lines were used for measuring ABP, sampling arterial blood gases, metabolites and cytokines. The fetal venous line was used for administration of fluids and post-operative antibiotics. Stainless steel electrodes were sewn onto the fetal chest to monitor ECG. A polyvinyl catheter was also placed in the maternal femoral vein. Stainless steel electrodes were additionally implanted biparietally on the dura for the recording of electrocorticogram, ECOG, as a measure of summated brain electrical activity [results reported elsewhere (16)]. An inflatable silicon rubber cuff (in vivo Metric, Healdsburg, CA) for UCO induction was placed around the proximal portion of the umbilical cord and secured to the abdominal skin. Once the fetus was returned to the uterus, a catheter was placed in the amniotic fluid cavity. Antibiotics were administered intravenously to the mother (0.2 g of trimethoprim and 1.2 g sulfadoxine, Schering Canada Inc., Pointe-Claire, Canada) and fetus and into the amniotic cavity (1 million IU penicillin G sodium, Pharmaceutical Partners of Canada, Richmond Hill, Canada). Amniotic fluid lost during surgery was replaced with warm saline. The uterus and abdominal wall incisions were sutured in layers and the catheters exteriorized through the maternal flank and secured to the back of the ewe in a plastic pouch. Postoperatively, animals were allowed 4 days to recover prior to experimentation and daily antibiotic administration was continued intravenously to the mother (0.2 g trimethoprim and 1.2 g sulfadoxine), into the fetal vein and the amniotic cavity (1 million IU penicillin G sodium, respectively). Arterial blood was sampled for evaluation of the fetal condition and catheters were flushed with heparinized saline to maintain patency. Animals were 130 ± 1 day GA on the first day of the experimental study.

Umbilical cord occlusion protocol. The experimental protocol has been reported (16, 20, 21). Briefly, all animals were studied over a ~6 h period. Fetal chronic hypoxia was defined as arterial O2Sat <55% as measured on postoperative days 1–3 and at baseline prior to beginning the UCOs. The first group comprised five fetuses that were also spontaneously hypoxic (n = 5, H/UCO). The second group of fetuses was normoxic (O₂Sat more than 55% before UCOs) (n = 9, N/UCO). As reported, after a 1–2 h baseline control period, the animals underwent mild, moderate, and severe series of repetitive UCOs by graduated inflation of the occluder cuff with a saline solution (16). During the first hour following the baseline period, mild variable FHR decelerations were performed with a partial UCO for 1 min duration every 2.5 min, with the goal of decreasing FHR by ~30 bpm, corresponding to a ~50% reduction in umbilical blood flow (22, 23). During the second hour, moderate variable FHR decelerations were performed with increased partial UCO for 1 min duration every 2.5 min with the goal of decreasing FHR by ~60 bpm, corresponding to a ~75% reduction in umbilical blood flow. Animals underwent severe variable FHR decelerations with complete UCO, i.e., ~100% reduction of umbilical blood flow, for 1 min duration every 2.5 min until the targeted fetal arterial pH of <7.00 was detected, at which point the repetitive UCO were terminated. A summary of timings is reported in Table 1. These animals were then allowed to recover for 48 h following the last UCO. Fetal arterial blood samples were drawn at baseline, at the end of the first UCO of each series (mild, moderate, severe), and at 20 min intervals (between UCO) throughout each of the UCO series, as well as at 1, 24, and 48 h of recovery. When pH <7.00 was measured, the UCO were stopped and this time point noted as the end of the occlusions. We then obtained the precise pH = 7.00 time point by linear interpolation from this last measured pH value. All blood samples were analyzed for blood gas values, pH, lactate and base excess (BE) with an ABL-725 blood gas analyzer (Radiometer Medical, Copenhagen, Denmark) with temperature corrected to 39.0°C. Plasma from the 4 ml blood samples was frozen and stored for cytokine analysis, reported elsewhere (24). After the 48 h recovery blood sample, the ewe and the fetus were killed by an overdose of barbiturate (30 mg sodium pentobarbital IV, MTC Pharmaceuticals, Cambridge, Canada). A post mortem was carried out during which fetal sex and weight were determined and the location and function of the umbilical occluder were confirmed. The fetal brain was perfusion-fixed and subsequently dissected and processed for later immunohistochemical study (25).

Data acquisition and pre-processing. A computerized data acquisition system was used to record fetal systemic arterial and amniotic pressures and the ECG signal (26). All signals were monitored continuously throughout the experiment. Arterial and amniotic pressures were measured using Statham pressure transducers (P23 ID; Gould Inc., Oxnard, CA). Fetal systemic ABP was determined as the difference between instantaneous values of arterial and amniotic pressures. A PowerLab system was used for data acquisition and analysis (Chart 5 For Windows, ADInstruments Pty Ltd, Castile Hill, Australia). Pressures, ECOG and ECG were recorded and digitized at 1,000 Hz for further study. For ECG, a 60 Hz notch filter was applied. R peaks of ECG were used to derive the heart rate variability (HRV) times series (26). The time series of R-R peak intervals were then uniformly resampled at 4 Hz (26). A representative FHR signal is shown in Figure 1A: a visual assessment of the whole FHR trace reveals that FHR variability increases when UCO strength is increased.

Metabolites data (pH, lactate and BE) is obtained by blood sampling performed at specific times during the experiment (vertical dashed lines in Figure 1B). In order to have metabolites data at any time, we assume a linear drift between two successive measurements and consequently perform a linear interpolation between two measurements times. We thus obtain a piece-wise linear time series sampled at 4 Hz, depicted in Figure 1B as a black curve. Using this interpolated data, as noted above, the time t_pH when pH = 7.00 is computed in each fetus as indicated in Table 1.

Fetal cardiovascular decompensation (CVD). CVD has been reported in detail in Frasch et al. (15, 16) and Gold et al. (27). The visual representation of CVD can be found in these publications, e.g., in the Figure 2 in Frasch et al. (16). The reader can readily observe the pronounced pathological hypotensive responses to the UCO-triggered FHR decelerations during the CVD. This behavior is in stark contrast to the normally observed ABP increases during the occlusions which compensate the hypotension caused by the FHR decelerations. As we reported, once this pattern conversion from hypertensive to hypotensive responses occurs, it persists until the UCOs are stopped. Its effects are also seen directly in the brain electrical activity (15, 16). It is hence easy to reliably visually identify the timing of the onset of CVD in each recording. Consequently, during UCOs, by expert visual inspection, we noted the individual time point t_CVD at which three successive hypotensive ABP responses to UCO-triggered FHR decelerations occurred. Quantitatively, with hypotensive ABP response we refer to the failure of ABP to rise during UCO-triggered FHR deceleration above the preceding baseline value when compared to the average ABP rise during the UCO series prior to the CVD. We refer to this animal-specific time point as the ABP sentinel corresponding to the timing of CVD. As an illustration, t_CVD is reported in Figure 1B as a vertical black line.

In section 4.1, the four features — computed in overlapping time-windows — evolution in time are firstly presented and studied with respect to their relations to UCO strength. Because these features are computed at a given time-scale τ, they offer a description of the FHR dynamics at this time-scale. We thus explore the correlation between the features at a given time-scale τ and the measured values of the metabolites — including the pH. This global analysis, presented in section 4.2, is performed using all available time-windows and all available animals. We then reduce the dimensionality of the analysis by averaging results over the long-term time-scales, as defined and presented in section 4.3. This allows us to examine more clearly how the features evolve jointly with the UCO strength for the entire cohort, while quantifying the variability between animals. We then examine quantitatively in section 4.4 how these long-term features correlate with metabolites. We then combine them in an appropriately normalized vector; we are then able to describe the large variability across the subjects in the population as the variability of this vector in the early stages of the experiments. This allows us to define a measure of the degradation of the health state of an animal as the distance from healthy state. Finally, we propose in section 4.5 to use this "individual" distance as a novel indicator — or sentinel — to alert for the degradation of the health status due to CVD. We also show that this indicator/sentinel matches very well with pH measurements.

Following achievements in adults and the seminal contribution in Akselrod et al. (41), frequency-based features were used to model linear temporal dynamics in FHR (42–46). To permit richer descriptions of the non-linear dynamics of FHR, information theoretic quantities were used such as entropy rates (34, 47–50), as well as several nonlinear transforms (45, 51–53), and scale-free or (multi)fractal paradigms (54–57). For overviews, interested readers are referred to, e.g., (9, 11, 58–64). An important limitation in the use of these features lies in their dependence on high quality fetal electrocardiogram (ECG) or magnetocardiogram (MCG) data as input. Such data are not readily available in the majority of clinical settings, with over 90% of North American hospitals, for example, still relying on CTG monitors during labor. CTG however provides FHR at a 4 Hz sampling rate, to be compared to 1,000 Hz sampling rate golden standard available with ECG or MCG, while vagally mediated HRV is found on a time scale that goes beyond what is captured at 4 Hz sampling rate. This results in information loss (19, 26, 65, 66). Beyond the mere design of features and their standalone use, numerous efforts were devoted to devise multiple-feature decision rules, often based on supervised learning and machine learning [cf., e.g., (9, 12, 62, 63, 67–69, 69–72)].

In the present work, four FHR features, whose definitions depend on the timescale, are computed on the whole FHR dataset derived from an animal model of human labor to quantify the evolution of FHR temporal dynamics. That means that in our approach we do not rely on considering UCO periods only, but are able to process the entire FHR signal as it would be available in real-time in clinical setting. These quantities are local statistical averages that probe the variation, the amplitude of fluctuations and the information content at a given time scale. They are purely statistical quantities that can be computed even when some data points are missing. Firstly, we qualitatively related the variations of such timescale-dependent quantities to the UCO strength; secondly, we quantitatively computed their correlation to metabolite and pH measurements.

As to the etiology of CVD, we propose a role for the Bezold-Jarisch reflex, a vagal cardiac depressor reflex, as part of a complex dynamic interplay, based on the observations of acidemia-triggered inflammation in fetal sheep (16, 17, 25), and studies in adult species linking rising systemic acidemia and inflammation with worsening cardiac contractility, impaired beta-adrenergic and potentiated bradycardic responses (25, 73–77). We suggest that the integrated ability of the four FHR features introduced in this study to track the individual evolution of acidemia and cardiovascular responses stems from capturing the individual complex interplay of the vagally mediated sensing of acidemia and the Bezold-Jarisch reflex, i.e., also vagally mediated intermittent hypotensive ABP responses to UCO-triggered FHR decelerations. This hypothesis needs to be validated in specifically designed animal experiments, for example by repeating the experiments underlying the present study with the variation of performing cervical bilateral, left or right vagotomies. This would allow evaluating the contribution of the vagus nerve to the dynamic interplay between the progressive systemic acidemia, the ensuing systemic inflammatory response, accounting for vagus nerve's lateral asymmetry, to the evolution of FHR decelerations and ABP responses over the period of worsening UCOs comparable in duration to stages 1 and 2 of pushing (18).

We show the relevance of timescales ranging in [2.5 − 8] seconds (equivalently [0.125 − 0.4] Hz in frequencies) for early detection of both acidemia and CVD, matching the scales classically used in FHR analysis and referred to as long-term (57, 63). We observed that reduced pH closely relates to larger m_LT which may be interpreted as an increase of baseline FHR (62, 63), and lower entropy rate h_LT, in agreement with earlier findings reported in the literature (34). More importantly, a per-individual distance metric was constructed from these four (population-normalized) features to quantify a self-referencing departure from a healthy state for each subject independently. Such a definition raises two issues. Firstly, it requires, as is often the case, that monitoring is started early enough while the fetus is still in a healthy condition, so as to create a self-reference to normal on a per-individual basis. If fetuses are already in distress when monitoring is initiated, the distance, albeit increasing with distress, may fail to detect CVD correctly. Secondly, the definition of the vector, and hence the distance, requires a normalization, which is performed in the present work at the population level, i.e., using an average across subjects. Although such an average should converge rapidly with the population size, this dependence requires further investigations.

It has been documented that sheep fetuses have an individual cardiovascular phenotype in their responses to increasing acidemia due to repetitive intermittent hypoxia (15). Chronically hypoxic fetuses have diminished cardiovascular defenses to hypotensive stress (78). Studying the same dataset, we demonstrated that under the conditions of repetitive UCOs and in comparison to the fetuses who were normoxic on the onset of the UCOs, the hypoxic fetuses exhibit accelerated acidosis (21), altered temporal profile of neuroinflammation following UCOs (20) and deceleration reserve (79). In the present study, we could not identify any influence of the phenotype (normoxic/hypoxic) in any of the metrics including the distance D. We are currently exploring whether the initial vector u→0 may contain such information. Conversely, the finding that the presented approach functions well without the consideration of a pre-existing hypoxia or pattern of labor contractions (UCO severity) is an additional bonus from the clinical viewpoint. Lastly, we recognize that the group of chronically hypoxic animals may have failed to recover from surgical instrumentation adequately, i.e., they were already decompensating rather than becoming “spontaneously” hypoxic for reasons of utero-placental dysfunction preceding the surgery.

Overall, the constructed distance proves able to detect accurately the occurrence of acidemia and CVD from the analysis of FHR only, and without recourse to pH. This opens the route to investigating the relevance of such metrics in clinical practice, as it is non-invasive and much faster than biochemical measurements like pH. Further, for practical purposes, the present studies show that the computation of features and distance is robust to FHR sampled at 4 Hz and, to some extent, to missing data. Also, the FHR features and distance are computed in sliding-windows, permitting an on-line and quasi-real time analysis of the evolution of the dynamics of FHR, and thus in relation to a local health state of the fetus. The extent to which the 20-min sliding-window size, chosen here for proof-of-concept developments, can be further reduced to 10 or 5-min, is under investigation. In conclusion, we propose a real-time FHR-based metric predicting CVD which should be of a great help for health practitioners managing the delivery.

The data analyzed in this study is subject to the following licenses/restrictions: experimental data may be made available on request (to MF). Requests to access these datasets should be directed to mfrasch@uw.edu.

The animal study was reviewed and approved by University of Western Ontario Council on Animal Care/Canadian Council on Animal Care.

SR designed the signal processing tools, ensured their practical implementation, conducted their application to data, and the analysis and interpretation of the results. He also prepared the figures and tables reported in the article. NBG contributed to the design of the signal processing tools, to the analysis and interpretation of the results, and to the writing of the manuscript. PA contributed to the interdisciplinary connections between signal processing and medical doctor teams, to the interpretation of the results, and to the writing of the manuscript. NG contributed to the measurements and the manuscript. MF designed the experiment, conducted the surgeries, and measurements. He also performed the expert visual detection of the cardiovascular decompensation used as ground truth here. He contributed to writing the article and to the interpretation of the results. All authors contributed to the article and approved the submitted version.

MF has an aECG patent WO2018160890A1. The remaining authors declare that the research was conducted in the absence of any other commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.