Electrohysterography extracted features dependency on anthropometric and pregnancy factors

The Uterine Electromyogram often referred as the Electrohysterogram (EHG) is a signal that has the potential to be used for pregnancy monitoring and preterm risk evaluation, amongst other applications. Classification algorithms success heavily rely on the adequacy of the selected features. In this study, Linear Mixed Models are used to evaluate a selected set of EHG features dependency on the gestational age, placental position, age, body mass index (BMI), gravidity and previous cesarean section. The obtained results point out, for example, to a dependency of some EHG features on the BMI and placental position, on the form of a low pass filtering effect. On the other hand, it was found that all the studied entropy features variation with gestational age was not statistically validated. Several conclusions can be drawn from the obtained results, which could be of interest for the increasing number of researchers using the EHG in machine learning computational algorithms, where the features are often selected without a disclosed criterion.