Computational Method for Preterm Labor Prediction using Electrohysterogram

Abstract

Electrohysterogram (EHG) is a noninvasive approach for recording the electrical activity of the uterine muscles (myometrium). It is also known as Uterine Electromyogram (Uterine-EMG). Since it is a non-invasive alternative, it is a safe and painless procedure to monitor uterine activity. In this paper, a wavelet transformation technique was used to preprocess the raw EHG signal, to remove artifacts present in it. A secondary time series was computed from the peak intervals of the EHG signals. This was followed by the extraction of the non-linear features from this time series. The classification of labor and non-labor signals was performed with a KNN algorithm. An accuracy of 90.32% was achieved with the KNN model. With supervised learning models, labor contractions were correctly identified in women who had just entered their third trimester but had not crossed the 37th week mark. With this knowledge, potential cases of preterm births could be identified. This knowledge could assist doctors in premeditating the prevention of still-births.