LaborEase An Artificial Intelligence-Based Wearable Electrohysterography Device for Preterm Pregnancy Detection

Preterm birth remains a leading cause of neonatal mortality and long-term health issues globally. Accurate early prediction can significantly improve maternal care and reduce risks to infants. This research presents a machine learning-based framework for the early detection of preterm birth using Electrohysterogram (EHG) signals. We developed a custom device to capture uterine electrical activity and combined the recorded signals with an open dataset (TPEHGDB) to enhance data volume and diversity.

From 1000-second EHG recordings, we extracted key features such as Count Contraction, Length of Contraction, Standard Deviation, Entropy, and Contraction Time. These features were chosen for their relevance in representing uterine dynamics. The target classification is binary: Preterm (1) or Term (0).

We implemented and evaluated multiple supervised machine learning models, including Random Forest, Decision Tree, Logistic Regression, K-Nearest Neighbors, Multilayer Perceptron (MLP), and Support Vector Machine (SVM). To address class imbalance, Synthetic Minority Over-sampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN) was applied. Performance was validated using 10-Fold Cross-Validation with metrics like Accuracy, Precision, Recall, F1-score, ROC-AUC, Specificity, Cohen’s Kappa, and Matthews Correlation Coefficient.

Results showed that Random Forest and  Descision tree models delivered superior predictive accuracy and robustness, demonstrating the effectiveness of using engineered signal features in preterm birth prediction. This study underscores the potential of integrating biomedical signal processing and machine learning to support timely clinical decisions in maternal health care.

Keywords: Electrohysterogram, Preterm Birth, Machine Learning, Random Forest, Descision tree , Feature Extraction, SMOTE-ENN, EHG Signal Analysis, Binary Classification, Maternal Health