Prediction of preterm deliveries from EHG signals using machine learning

Area of Science:

• Biomedical Engineering
• Computational Medicine
• Obstetrics and Gynecology

Background:

• Preterm birth rates are increasing globally, posing significant health risks to infants, including long-term developmental issues and increased healthcare costs.
• Current methods for predicting preterm birth are largely subjective and lack accuracy.
• Electrohysterography (EHG), the analysis of uterine electrical signals, shows potential for diagnosing labor and predicting preterm delivery.

Purpose of the Study:

• To explore the utilization of Electrohysterography (EHG) techniques for earlier prediction of preterm delivery.
• To develop and evaluate a supervised machine learning approach for classifying term and preterm birth records.

Main Methods:

• A supervised machine learning model was developed using an open-source dataset of 300 uterine electrical signal records (38 preterm, 262 term).
• Synthetic Minority Oversampling Technique (SMOTE) was applied to address the class imbalance of preterm births.
• Cross-validation techniques were employed to evaluate the model's performance against existing studies.

Main Results:

• The developed machine learning approach achieved high accuracy in classifying preterm and term births.
• Performance metrics include 96% sensitivity, 90% specificity, and a 95% area under the curve (AUC).
• The model demonstrated a global error rate of 8% using a polynomial classifier.

Conclusions:

• The study demonstrates that Electrohysterography (EHG) combined with machine learning can effectively predict preterm birth earlier in pregnancy.
• This approach shows improved performance compared to existing studies, offering a more objective and accurate prediction method.
• Early prediction of preterm birth can lead to timely interventions, potentially improving infant health outcomes and reducing societal economic burden.