Deep Learning Model Using Continuous Skin Temperature Data Predicts Labor Onset.

Area of Science:

• Reproductive endocrinology and obstetrics
• Biomedical engineering
• Computational biology

Background:

• Mammalian labor onset is often preceded by body temperature changes, but this has not been studied in humans.
• Pregnancy monitoring lacks precise methods for predicting labor onset.

Approach:

• Utilized wearable smart rings to collect continuous skin temperature data from 91 pregnant women.
• Analyzed daily steroid hormone samples and their relationship with body temperature trends in 28 pregnancies.
• Developed a novel autoencoder long-short-term-memory (AE-LSTM) deep learning model for daily labor onset prediction.

Key Points:

• Skin temperature patterns correlated with urinary hormones and labor type (spontaneous vs. induced/Cesarean).
• Spontaneous labors showed distinct hormonal profiles (higher estriol to α-pregnanediol ratio) and more stable circadian rhythms.
• The AE-LSTM model predicted labor onset with an average error below 2 days at 8 days prior, achieving 79% accuracy within a 4.6-day window 7 days before labor.

Conclusions:

• Continuous skin temperature monitoring provides insights into labor progression and hormonal status during pregnancy.
• Deep learning analysis of temperature data may yield valuable clinical tools for enhanced pregnancy management and labor prediction.