By studying the electrohysterogram (EHG) signals, which reveal the changes of the
electric potentials of the uterus, we can tackle the problem of preterm birth
prediction non-invasively. The aim is to separate the sets of preterm and term delivery
EHG records based on the estimation of the velocity and direction of
the electric potential propagation. We calculate the directions and velocities
using the short-time cross-correlation procedure. We compare the values of pairs
of parallel signals above and below, and left and right from the navel, and determine
the time delays of the pairs of signals. Sequences of velocities and directions of
travel of the propagation waves are calculated next. We defined features
that describe the direction and velocity of propagation, and
their regularity, and added some features that describe original signals. The
features allow us to better understand the electrical activities of the uterus.
We perform separation of the groups of records inside
the selected frequency bands, and study the impact of different physiological
mechanisms. In the frequency band below 1.0 Hz, the most influent on the electrical
activity are the uterine contractions, while above 1.0 Hz, exclusively mother's heart
beating. We analyse the records as whole, and characterize the selected individual
intervals of records, i.e., intervals containing contractions, or non-contraction
intervals. Besides the type of the intervals and the selected frequency
band, we estimate also the classification accuracy of the defined features with regard
to the time of recording. We conclude that studying the influence of the maternal heart
is of key importance during separating the groups of records since the highest
classification accuracies were obtained above 1.0 Hz. The maternal heart influences
are the most contributive when classifying non-contractions intervals. The highest
classification results were obtained using those features that describe regularity
of the original signals, and frequency of propagation of the electric potentials in
horizontal direction. The last feature appeared to be the most important feature
when separating records recorded early during pregnancy, thus allowing early prediction
of preterm birth in clinical practice.
|