ABSTRACT
The purpose of
this project was to develop a noninvasive method to detect fetal heart
activity in utero. Developing such a method is of high importance, since
most of the current methods suffer from a high percentage of incorrect
diagnoses: false alarms resulting in operational deliveries, and a small
but significant percentage of undetected fetal distress.
BACKGROUND
ECG (Electrocardiogram)
is a signal, produced by the electrical activity of the heart. It is measured
between two electrodes attached to the skin. The ECG has a known shape,
and clinical diagnoses can be made by its monitoring. The abdominal ECG
measured from a pregnant woman contains the maternal signal, the fetal
signal, and additive noise. The noise is mainly due to the muscles electrical
activity - EMG. The fetal ECG signal has an amplitude 10-100 times smaller
than the maternal. Currently a Doppler Ultrasound device is used for monitoring
fetal heart rate. Our goal was to use the Doppler signal for detection
of the fetal ECG.
BASIC APPROACH
The Doppler signal
is filtered to produce envelopes from the bursts. Adjacent envelopes are
cross-correlated to give the interval between fetal heart beats. Using
these intervals, it is possible to average a fetal ECG template, starting
the averaging from an arbitrary point, and then fine tuning it. By searching
above a threshold, the maternal beats are identified, and averaged to a
template. The additional noise is modeled as an autoregressive stochastic
process, using noise only segments. The model we use to reconstruct each
fetal ECG complex, is based on a model used by Lange et. al. for
brain signals - EEG. The block diagram of the model is shown in Fig. 1.
The filters coefficients
are found using the Least Squares solution. Then, the fetal complex shape
is given by the output of its filter.
This algorithm was
realized in a Graphical User Interface. The output of the interface is
shown in the Fig. 2. In the reconstructed fetal ECG signal, the dotted
line is the measured abdominal ECG, and the continuous line is the reconstructed
one.
TOOLS
The signals were
sampled with a Data Translation A/D card. The simulations and the
Graphical User Interface were programmed in MATLAB for Windows,
on a PC.
CONCLUSION
The algorithms were found
to be efficient, and the results were satisfactory. The developed interface
gave excellent results in several patients tested so far, and in several
situations. The dependence of the algorithm on the Doppler signal, requires
special attention in measuring it, and making sure that its quality is
sufficient. The interface is a tool for achieving good results fairly easily
from the measured signals.
ACKNOWLEDGMENTS
We would like to
thank our supervisor Danny Lange for his guidance throughout the
project, and Dr. Reuven Lewinsky for providing the signals, and
for giving us insight into the problem. We would also like to thank the
laboratory staff for the technical support.
