CHN-ZZEHG: A Dataset of Uterine Electromyography Signals from Pregnant Women for Studying Human Labor Mechanisms and Preterm Birth Prediction

Background: Preterm birth is a leading cause of neonatal mortality and long-term health issues, yet clinical practice still lacks effective risk prediction tools. Electrohysterography (EHG), as a non-invasive and objective technique for monitoring uterine activity, shows promise in elucidating the mechanisms of labor onset and predicting preterm birth. Objective: To construct a large-scale, standardized public dataset, named CHN-ZZEHG, of uterine electromyography signals from pregnant women and to explore the feasibility of preterm birth prediction models based on this dataset. Methods: Recruitment of prospective 400 pregnant women was conducted from December 2024 to December 2025 at the Guangzhou Women and Children's Medical Center, affiliated with Guangzhou Medical University. A total of 444 high-quality EHG recordings were collected, covering various delivery outcomes such as spontaneous term labor, term non-labor, preterm birth, and medically indicated termination of pregnancy. All data were acquired following a unified acquisition protocol using a standardized four-channel abdominal electrode configuration and underwent preprocessing with 0.1–3 Hz bandpass filtering and 50 Hz notch filtering. Each recording was accompanied by detailed clinical metadata, including gestational age, mode of delivery, and medication use. Based on this, a support vector machine algorithm was used to construct a preterm birth prediction model. Results: Technical validation demonstrates the high reliability and scientific quality of this dataset: signal quality assessment indicated that over 92% of recordings achieved a "good" or higher rating. Physiological plausibility analysis revealed that with increasing gestational age, the average burst duration (DUR), power spectral density (PSD), and median frequency (MF) of EHG signals significantly increased, while sample entropy (SampEn) significantly decreased (P < 0.01). A preliminary preterm birth prediction model built on the CHN-ZZEHG dataset achieved an accuracy of 84.5%, precision of 82.1%, recall of 79.8%, and an F1-score of 0.81, validating the dataset's direct application value in prediction algorithm development. Conclusion: The CHN-ZZEHG dataset constitutes a large-scale, standardized, and clinically richly annotated benchmark resource, providing crucial support for in-depth exploration of human labor mechanisms and the development of novel preterm birth prediction methods. This dataset has been deposited in the PhysioNet open database and is publicly available under the CC-BY 4.0 license.研究背景：早产是新生儿死亡与长期健康问题的首要诱因，但当前临床实践仍缺乏有效的风险预测工具。子宫肌电图（Electrohysterography, EHG）作为一种无创且客观的子宫活动监测技术，在阐明分娩启动机制、预测早产方面展现出应用潜力。 研究目标：构建一款大规模、标准化的孕妇子宫肌电信号公开数据集CHN-ZZEHG，并探索基于该数据集构建早产预测模型的可行性。 研究方法：本研究于2024年12月至2025年12月在广州医科大学附属广州妇女儿童医疗中心开展，前瞻性招募400名孕妇。最终共收集到444条高质量EHG记录，涵盖自发性足月分娩、足月未分娩、早产以及医学指征终止妊娠等多种分娩结局。所有数据均按照统一采集方案，采用标准化四通道腹部电极配置进行采集，并经过0.1~3Hz带通滤波与50Hz陷波滤波的预处理步骤。每条记录均附带详细的临床元数据，涵盖孕周、分娩方式与用药史等信息。基于该数据集，本研究采用支持向量机（Support Vector Machine, SVM）算法构建早产预测模型。 研究结果：技术验证结果表明本数据集具有较高的可靠性与科学质量：信号质量评估显示，超过92%的记录达到“良好”及以上评级。生理合理性分析显示，随着孕周增加，EHG信号的平均爆发持续时长（burst duration, DUR）、功率谱密度（power spectral density, PSD）与中位频率（median frequency, MF）均显著升高，而样本熵（sample entropy, SampEn）则显著降低（P < 0.01）。基于CHN-ZZEHG数据集构建的初步早产预测模型达到了84.5%的准确率、82.1%的精确率、79.8%的召回率以及0.81的F1值，验证了该数据集在预测算法开发中的直接应用价值。 结论：CHN-ZZEHG数据集是一款大规模、标准化且临床注释丰富的基准资源，为深入探究人类分娩机制与开发新型早产预测方法提供了关键支撑。该数据集已存档于PhysioNet开放数据库，并采用CC-BY 4.0协议公开获取。