Residual-Noise-Aware Uterine Contraction Detection From Electrohysterography Using Bayesian Tensor Decomposition

This is the authors' abstract. We don't add key points for this paper.

Muscle cell electrical activity serves as the fundamental trigger for mechanical force generation, underpinning its widespread applicability in clinical practice. Electrohysterography (EHG) offers a non-invasive alternative for uterine contraction monitoring. However, EHG signals are often contaminated by physiological interference from adjacent organs, leading to high false-positive rates in contraction detection. To address these challenges, we propose a residual noise aware uterine contraction detection framework that combines Bayesian tensor decomposition with channel optimization. The method consists of three key components: 1) a signal separation module based on Bayesian tensor decomposition, which decomposes EHG recordings into contraction-related activity, distributed physiological interference, and residual noise; 2) a residual-noise guided channel optimization module that identifies the most representative channel for contraction reconstruction; and 3) an automatic contraction recognition algorithm that extracts precise contraction events from the reconstructed curves. Experimental results on public databases demonstrate that the proposed framework achieves superior contraction detection performance compared with existing approaches, significantly improving sensitivity and F1-score while reducing false detections. Ablation studies further reveal that tensor decomposition and residual-noise-guided channel optimization provide complementary benefits. These findings suggest that the proposed framework offers an effective and computationally efficient solution for EHG-based uterine contraction monitoring and may facilitate future applications in preterm birth assessment and labor management.