TL;DR
This paper introduces a novel method to characterize uterine contractions by analyzing the synchrony and coherence of electrical signals (EHG) recorded from multiple channels. By combining EHG with traditional TOCO signals, they achieve high accuracy in predicting imminent labor, offering a more insightful approach than current methods.
What Problem Does This Paper Address?
Predicting the onset of labor and identifying high-risk pregnancies, particularly preterm birth, is a critical challenge in obstetrics. Current methods like the tocodynamometer (TOCO) primarily measure the physical force of contractions, providing limited information about their underlying electrical activity, synchrony, or coherence. This lack of detailed information can lead to unnecessary hospitalizations for women at low risk and missed opportunities for intervention for those at high risk. While electrohysterography (EHG) captures the electrical signals of uterine contractions, existing methods often struggle with accuracy and lack a clear clinical interpretation. This paper aims to bridge this gap by developing a method that leverages multi-channel EHG to quantify the spatial-temporal characteristics of uterine contractions, thereby improving the prediction of imminent labor.
Methods
The core of this work lies in a novel approach to characterize uterine contractions by analyzing multi-channel EHG signals. The process involves:
- TOCO-guided EHG Segmentation: Contractile time windows are first identified from TOCO signals, which are considered reliable for marking contraction onset and duration. These windows are then used to segment the simultaneously recorded multi-channel EHG signals. This ensures that the analysis focuses on actual contractile events, avoiding artifacts.
- Spatial Regularity Characterization (SamEn): For each segmented EHG signal, the authors compute Sample Entropy (SamEn). This metric quantifies the regularity of the electrical activity across different spatial locations (channels) within the uterine wall during a contraction. A lower SamEn value indicates more regular, synchronized electrical patterns.
- Synchrony Characterization (CCC): To measure how synchronized the contractions are across the entire uterus, the Concordance Correlation Coefficient (CCC) is employed. This metric assesses the agreement between EHG signals from different channels. A higher CCC value (approaching 1) signifies greater synchrony.
- Feature Combination and Classification: The derived SamEn and CCC values serve as features to classify patients into "labor" (birth within T hours) and "pregnancy" (birth after T hours) groups.
Datasets
The study utilizes the Icelandic 16-electrode Electrohysterogram Database from PhysioNet. This database contains 122 EHG recordings from 45 participants, with simultaneously recorded TOCO signals. The recordings were collected at Landspitali University Hospital and Akureyri Hospital. The data was categorized based on the time-to-delivery (TTD), with critical thresholds of 24 hours, 72 hours, and 120 hours used for analysis.
Key Insights and Ideas
- Synchrony and Coherence are Key: The paper strongly suggests that the synchrony and coherence of uterine contractions, not just their intensity, are crucial indicators for predicting labor onset.
- Multi-channel EHG Offers Superior Information: By analyzing EHG signals from multiple channels, the method captures the spatial-temporal dynamics of contractions, providing richer information than single-channel EHG or TOCO alone.
- TOCO as a Reliable Anchor: Using TOCO signals to identify contractile time windows is a clever way to leverage its reliability in marking contraction events while still extracting detailed information from EHG.
- Entropy as a Measure of Regularity: Sample Entropy proves to be an effective metric for quantifying the regularity of electrical activity, which decreases as labor approaches.
- Predictive Power Extends to Early Stages: The developed features demonstrate significant predictive power even when measurements are taken up to 120 hours before delivery, highlighting their potential for early risk assessment.
Key Results
The proposed method achieved impressive results in discriminating imminent labor:
- High Accuracy for Imminent Labor (≤ 24h): The SamEn feature achieved an Area Under the Curve (AUC) of 0.834, with a sensitivity of 0.909, specificity of 0.712, and accuracy of 0.743.
- Sustained Predictive Power: Even when predicting labor 120 hours in advance, the SamEn feature maintained a respectable AUC of 0.698.
- Comparison to SOTA: The study's results, particularly for SamEn, generally outperform other state-of-the-art methods reported in the literature, especially when considering the direct clinical applicability without complex machine learning models.
- Number of Channels Matters: The discriminatory power of SamEn generally increases with the number of EHG channels used, suggesting that capturing more spatial information is beneficial.
Limitations and Future Work
- Small Dataset Size: The authors acknowledge that the validation was performed on a relatively small dataset, with only 14 patients giving birth within 24 hours. This limitation could lead to fluctuations in performance scores.
- Spontaneous vs. Induced Labor: The study did not differentiate between spontaneous and induced labor, relying solely on the time difference between gestational age at recording and delivery.
- Future Work: The authors encourage clinical trials with larger patient cohorts to further validate their findings. They also suggest that simultaneously recording both uterine electrical activity (EHG) and contractions (TOCO) offers a superior approach for evaluating uterine contractions compared to current monitoring devices. Further research could explore the application of these methods in more diverse clinical scenarios.
Summary updated