Introduction: Throughout pregnancy fetal health is monitored regularly to ensure proper care. Fetal heart rate (fHR) is a vital physiological marker for assessing the cardiovascular and autonomic nervous system.
Methods: In this study, fHR was extracted from fetal phonocardiography (fPCG) and fetal abdominal electrocardiography (fECG) signals obtained from 100 singleton pregnancies (gestational age (GA) 27-40 weeks), in a real-world clinical setting. Microphones recorded the PCG and were placed bilaterally on the lower abdomen. Four ECG leads crossing the abdomen diagonally and laterally recorded the fECG. Open-source extraction algorithms were used to extract fetal heart sounds (fHS) and R-peaks from the PCG and ECG, respectively. A pretrained (on adults) duration dependent hidden Markov model was used to predict fHS after adjusting parameters to fit characteristics of fHS, and an algorithm based on independent component analysis was used to detect R-peaks. A 3-second moving average was calculated of interbeat intervals (IBI) obtained from S1, S2 and R estimates, as well as a median estimate of the three. To utilize all modalities, even in case of dropouts, a combined IBI annotation, from all three sources, was used to compare the four estimates using root mean square error (RMSE).
Results: Overall median RMSE decreased from ~31 ms from the fPCG estimates to 24 ms with all modalities combined. However, the fECG estimate had higher RMSE (41 ms). fPCG was superior to fECG, potentially due to the wide range of GA overlapping the period of low conductivity from fECG caused by the presence of vernix caseosa. The figure also shows how RMSE decreases for fECG when the vernix caseosa subsides in GA>37.
Conclusion: Our results highlight the potential of fPCG as a method for fHR monitoring during the entire third trimester.