Fetal ECG is the leading tool to analyze and detect abnormalities during fetal development. For example, to
identify fetal distress and intrauterine hypoxia. Currently, the most common method to monitor a fetus, especially
during labor is through fetal monitoring, which is based on auscultation, though there are more accurate methods,
based on the electrical signal of the fetus (the fetal ECG or FECG). There are currently two methods for obtaining
FECG. One is an invasive scalp electrode technique that is invasive and used during labor, while the other is
based on electrode patches that are placed on the monther’s abdomen to collect the signals. However, the signals
from the maternal abdominal are highly noisy and include the mixture of maternal ECG and a weak FECG on top
of maternal breath sound and increased level of noise. In practice the extraction of the FECG is done using long
sampling cycles, with high computational cost and with limited success which depends on the magnitude of the
fetal source. This is the most accurate method of objective fetal monitoring currently available, however, all
electrical methods of acquiring a reliable fetal signal have not proven to be accurate.
In this work we propose a new approach to extract the FECG from the joined maternal-fetal data by sampling the
joined ECG from at-least two sources and training a deep learning model to split the two in real-time. Since this
is a hard task, we present a patient specific tailored personalized self-supervised network. This method is trained
per patient (maternal and fetal) and can extract the maternal ECG in real-time and with high-quality. Based on
preliminary results on real data, we demonstrate that the hypothesis and solution are applicable and feasible.
The Fetal monitoring market is increasing rapidly and accelerated after the recent pandemic. According to the
insight partners market review, we accept to see a rise from $3.8b 2021 to $6.4b by 2028.