Preterm Delivery Prediction by Analysis of Multi-channel Uterine EMG signals

Abstract

Predicting the premature delivery is very important in preventing infant deaths and consequent health risks globally. The uterine Electromyography signals Electrohysterogram (EHG) signals, has been very promising and can prove to be a marker in diagnosing Preterm birth. In this study, the TPEHG DB (Term-Preterm Electrohysterogram Database) dataset with 300 records (262 term and 38 preterm records) are used. The raw uterine EMG signal is filtered and the linear, non-linear and statistical features are extracted. The extracted features are applied to machine learning classifiers. To improve the classification accuracy, Bayesian Hyperparameter Optimization technique is employed. Support vector machine (SVM) classifier with Bayesian Hyperparameter Optimization technique, tested using 10-fold cross-validation on 38 preterm records provided 96.667% accuracy.