Estimating uterine activity from electrohysterogram measurements via statistical tensor decomposition

TL;DR

This paper introduces a Bayesian tensor decomposition method to improve the analysis of electrohysterogram signals, enabling better separation of uterine activity from noise and interference, which could enhance pregnancy monitoring.

Contribution

The paper presents a novel statistical tensor decomposition approach for separating localized uterine activity from distributed signals in EHG measurements, outperforming existing methods.

Findings

More accurate estimation of uterine activity in simulated data.

Enhanced separation of EHG bursts from interference in real data.

Higher signal-to-noise ratios compared to alternative methods.

Abstract

Complications during pregnancy and labor are common and can be especially detrimental in populations with limited access to healthcare. A promising technology to address these complications is the electrohysterogram (EHG), which measures abdominal electric potentials. Since EHG recordings are measured noninvasively, they record uterine electrical activity together with activity from other interfering sources, making their analysis more challenging. To facilitate the analysis of EHGs, we separate these measurements into uterine activity that is more variable across different electrodes and over time, which we term localized activity, and activity that is more evenly distributed in space and time. We represent multi-electrode EHGs as tensors and develop a Bayesian tensor decomposition for estimating localized and distributed electrical activities. To demonstrate that our method can estimate localized and distributed activities more accurately than existing methods, we simulate EHG measurements based on previous characterizations of the activities comprising these measurements. Furthermore, we demonstrate the effectiveness of our method in separating EHG bursts from other interfering activities recorded in EHGs using real measurements from two public datasets. Our results show that our method denoises EHG bursts with higher signal-to-noise ratios, defined as power ratios between EHG bursts and segments with only baseline EHG activity, than alternative methods. Such accurate estimation of uterine and other activities from EHG measurements could be useful for more precisely characterizing patterns of activity in EHGs, which is a step forward towards developing reliable, portable devices to reduce risks during pregnancy and labor.