Simone Rahm

Simone Rahm

Master's Thesis

Fetal Heart Rate Extraction from Wearable Non-Invasive Electrocardiogram Devices


Katharina Jäger (M.Sc.), Dr. Heike LeutheuserProf. Dr. B. Eskofier


02 / 2021 – 08 / 2021


Nearly one third of all birth defects are congenital heart conditions. The global prevalence at birth is
estimated at almost 1.8 cases per 100 live births. In countrieswith a high Socio-Demographic Index (SDI),
congenital heart defects are among the top five causes of infant mortality, and among the top eight causes
in all countries. Therefore, an early evaluation of the fetal heart is essential for identifying fetuses at risk
and provide early care [3, 7].
In current clinical practice, the cardiac health status of the fetus is examined using cardiotocography
(CTG). This technology uses an ultrasound transducer attached to the maternal abdomen to measure
fetal heart rate. Simultaneously, uterine activity is captured by a pressure-sensitive transducer. However,
this technology has certain limitations.On one hand, current devices can only estimate the time-averaged
fetal heart rate due to the usage of window functions. On the other hand, CTG must be used by medical
professionals since an accurate placement of the ultrasound sensor is essential to obtain a high signal quality
[2, 5].
In contrast, non-invasive fetal electrocardiography (NI-fECG) is a measurement technique that has
the potential to address these challenges. This method records the electrophysiological signal on the skin
surface of the maternal abdomen using multiple electrodes. The measured signal is composed of the fetal
ECG, the maternal ECG and other interfering signals such as muscular or motion artifacts. Signal processing
methods can then be used to extract the fetal heart rate from the signal. In this way it is possible to
achieve a beat-to-beat measurement of the fetal heart rate, which provides additional diagnostic information
about heart rate variability [4, 6]. Furthermore, it is more feasible to integrate NI-fECG measurement
technology into wearable devices to enable a home-monitoring outside of the clinical environment [1].
This facilitates pregnantwomen’s access to prenatal care, especially in rural areas, as no professional staff
is needed to operate the device [5].
One reason why NI-fECG is not yet established in clinical practice is the lack of a large database with
uniformly recorded signals on which newalgorithms can be tested and compared. Recently, however, several
open-source datasets have been published that include recordings with different electrode placements
and fetal R-peak annotations from medical experts for reference [4].
Multiple signal processing algorithms have already been developed for fetal heart rate extraction from
NI-fECG. They can be divided into algorithms that consider only abdominal electrodes and algorithms
that require an additional electrode on the mother’s chest to record maternal ECG. When the maternal
ECG is available, it is assumed that components of the maternal ECG are present in both chest and abdominal
electrodes. Thus, the maternal ECG can be removed from the abdominal signals using adaptive
techniques, leaving the fetalECG. If only abdominal electrodes are available, the maternalECGis extracted
from the abdominal signals using either a template in the time domain, or blind source separation techniques
that operate in the spatial domain. Literature suggests that a combination of different algorithms,
so-called hybrid methods, may provide the best results [4].
However, to date, no objective comparison between existing fetal heart rate extraction algorithms has
been conducted. Therefore, there is a need to benchmark the existing algorithms on the various publicly
available datasets. The aim of this thesis is to develop a newhybrid algorithm for fetal heart rate extraction
from NI-fECG records, which should especially focus on signals measured with wearable devices. For
this purpose, the existing algorithms are compared and their strengths and weaknesses are identified in
order to achieve a suitable combination of the methods. The resulting approach will be tested on recordings
from existing open-access fECG databases as well as on an additional dataset that will be acquired
within a clinical study using a wearable device. In a final step, the developed algorithm is benchmarked
against state-of-the-art approaches.



] J. Dunn, R. Runge, and M. Snyder. Wearables and the medical revolution. Personalized Medicine, 15, 09
[2] P. Hamelmann, R. Vullings, A. F. Kolen, J.W. M. Bergmans, J. O. E. H. van Laar, P. Tortoli, and M. Mischi.
Doppler ultrasound technology for fetal heart rate monitoring: A review. IEEE Transactions on
Ultrasonics, Ferroelectrics, and Frequency Control, 67(2):226–238, 2020.
[3] E. Hernandez-Andrade, M. Patwardhan, M. Cruz-Lemini, and S. Luewan. Early evaluation of the fetal
heart. Fetal Diagnosis and Therapy, 42:161 – 173, 2017.
[4] R. Kahankova, R. Martinek, R. Jaros, K. Behbehani, A. Matonia, M. Jezewski, and J. Behar. A review of
signal processing techniques for non-invasive fetal electrocardiography. IEEE Reviews in Biomedical
Engineering, PP:1–1, 08 2019.
[5] M. Mhajna, N. Schwartz, L. Levit-Rosen, S.Warsof, M. Lipschuetz, M. Jakobs, J. Rychik, C. Sohn, and
S.Yagel.Wireless, remote solution forhomefetal andmaternal heart rate monitoring. American journal
of obstetrics and gynecology, page 100101, 03 2020.
[6] R. Sameni and G. Clifford. A reviewof fetal ecg signal processing issues and promising directions. The
open pacing, electrophysiology & therapy journal, 3:4–20, 01 2010.
[7] M. Zimmerman, A. Smith, C. Sable, M. Echko, L. Wilner, H. Olsen, H. Tasew, A. Awasthi, Z. Bhutta,
J. Boucher, F. Castro, P. Cortesi, M. Dubey, F. Fischer, S. Hamidi, S. Hay, C. Hoang, C. Hugo-Hamman,
K. Jenkins, and N. Kassebaum. Global, regional, and national burden of congenital heart disease, 1990
– 2017: a systematic analysis for the global burden of disease study 2017. 4:185–200, 01 2020.