In recent years, electrohysterography (EHG) signals have emerged as a promising research direction for predicting labor and diagnosing preterm birth due to their ability to reflect uterine contraction characteristics. Traditional intrauterine pressure catheter methods effectively measure contraction intensity but are invasive and pose a risk of infection, whereas EHG provides a safer, non-invasive alternative. However, current single-channel EHG analysis methods face limitations in capturing the complex spatiotemporal characteristics of uterine contractions.

This paper proposes a novel approach that utilizes graph networks to analyze multi-channel EHG signals, enabling the extraction of more comprehensive inter-channel correlation information to distinguish between pregnant and laboring states. Compared to single-channel analysis, multi-channel EHG signals provide a more detailed representation of the complex spatiotemporal dynamics of contractions. In this study, we calculated six functional connectivity metrics for each pair of EHG signals, constructing a graph network where electrodes served as nodes and connectivity values represented edges. Utilizing graph theory, we extracted four network features to characterize signal propagation and assess myocyte excitability. These features were then used to classify contraction signal segments into pregnancy and labor states.

The proposed graph network approach based on pairwise phase coherence outperformed traditional univariate analysis methods. Combining network features with univariate features further improved classification performance, achieving a sensitivity of 87.62%, specificity of 92.26%, and accuracy of 90.91%.

The proposed method effectively utilizes multi-channel EHG information to enhance the classification of pregnancy and labor states. This approach provides valuable insights for labor detection and preterm birth diagnosis, demonstrating the potential of graph network-based EHG analysis.