Machine Learning Techniques for the Prediction of Preterm Birth Using Electrohysterography Signals

Abstract

Preterm birth (PTB) is one of the dangerous pregnancy complications where the delivery occurs prior to 37 weeks of gestation. Prediction of PTB has been challenging since there are no objective measures to determine the uterine contractility. Electrohysterography (EHG) is a promising non-invasive approach which is currently being considered for improvement of external uterine monitoring. Automated analysis of EHG signals is a prospective method for monitoring pregnancy, detecting labor, and predicting PTB. This chapter focuses on the analysis of EHG signals and its application for the early diagnosis of PTB using machine learning techniques. Various public datasets and the techniques to extract appropriate information from the EHG signals are discussed. Analysis is done on the issue of class imbalance brought on by the relatively limited amount of PTB data as well as the efficacy of machine learning methods for early diagnosis of PTB. Further, potential research directions for EHG analysis and its application to PTB prediction are identified.