Towards Non-invasive Labour Detection: A Free-Living Evaluation

Area of Science:

• Biomedical Engineering
• Maternal-Fetal Medicine
• Machine Learning in Healthcare

Background:

• Detecting labor non-invasively during pregnancy presents challenges due to unsupervised, free-living data collection.
• Existing methods may require invasive procedures or controlled environments, limiting real-world application.

Purpose of the Study:

• To develop and validate machine learning models for non-invasive labor detection using physiological data collected in free-living conditions.
• To assess the feasibility of identifying high-quality data and physiological changes indicative of approaching labor.

Main Methods:

• Collected electrohysterography (EHG), heart rate (HR), and gestational age (GA) data in laboratory settings for model development.
• Simulated daily living activities to elicit artifacts and develop models for free-living data.
• Deployed machine learning models longitudinally in 142 pregnant women (22 weeks gestation to delivery), collecting 1014 hours of data.

Main Results:

• Machine learning models successfully processed unsupervised, free-living physiological data.
• The probability of labor was consistently higher in recordings from the last 24 hours of pregnancy compared to earlier weeks.
• Non-invasive detection of labor using a single abdominal sensor showed promising results.

Conclusions:

• Non-invasive labor detection from physiological data acquired in free-living conditions is feasible.
• Machine learning models can effectively handle challenges of unsupervised data collection for labor prediction.
• This approach holds potential for improved remote monitoring and management of pregnancy.