Non-Invasive Fetal Monitoring using Non-Contact Electrodes and Recurrent Neural Networks
Typ
Examensarbete för masterexamen
Program
Publicerad
2019
Författare
Annér, Albin
Kastö, David
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The fetal well-being is routinely monitored using a cardiotocograph (CTG), a combination
of a Doppler sensor used to measure the fetal heart beats and a pressure
sensor to measure uterine muscle contractions. However, the equipment is expensive
and there is a recurrent issue where the CTG confuses the maternal heart rate
for the fetal heart rate, leading to ambiguities that have resulted in adverese fetal
outcomes on multiple occasions. Non-invasive recordings of the fetal heart activity
on the maternal abdomen could constitute a viable alternative to Doppler ultrasound
recording. However, the potential sensed by an abdominal electrode is the
combination of many sources, making it an onerous task to extract the components
stemming from the fetal heart beats. The most difficult subsignal to circumvent is
the contribution from the maternal heart.
This thesis explores the viability of using non-contact electrodes to reliably measure
the fetal electrocardiogram, fECG, and an electrode is successfully designed for this
purpose. This electrode could be implemented in an array embedded in a piece
of clothing for a practical implementation of many uncorrelated electrodes, which
normally facilitates the signal separation process and improves its accuracy.
Separately, using the Non-Invasive Fetal Electrocardiogram (NI-fECG) database,
source separation and fetal heart rate extraction algorithms are evaluated, developed,
and improved upon. It is shown that using said algorithms in combination
with a novel fetal heart rate detector, fetal heart beats can be detected in a reliable
manner on the evaluated data.