Detection of preterm labor by partitioning and clustering the EHG signal

This study aims to predict the risk of preterm labor by analyzing electrohysterogram (EHG). To this purpose, the Term-Preterm EHG Database (TPEHG Database) with 300 EHG signals from pregnant women that are categorized into two classes of term (262) and preterm (38) has been taken into account. This research proposes an algorithm based on time-frequency analysis and thresholding methods for quantitative estimation of uterine contractions. This estimation dismantles the EHG signals into small segments where each segment refers to an event. Then, Linear Predictive Coding (LPC) is applied to extract features from these segments. This study also presents a new approach for classification of term and preterm signal records. To this end, the events were clustered using an unsupervised clustering method and then each cluster was categorized independently to detect term and preterm births. As a result, it was possible to omit the unrelated segments of each record using this approach. The results indicate a significant improvement in separability and accuracy in the preterm birth detection index.