Birth Defects Res. 2026 Mar;118(3):e70038. doi: 10.1002/bdr2.70038.
ABSTRACT
BACKGROUND: Pregnancy-related syndromes, such as hypertensive disorders, gestational diabetes mellitus, and preterm birth, pose a significant global health burden, affecting maternal and fetal outcomes. Traditional screening methods, reliant on isolated biomarkers or linear models, often fail to address the complex pathophysiology of these conditions.
METHOD: This review synthesizes current literature on machine learning applications in obstetric care, analyzing multimodal data integration from electronic health records, biochemical markers, multi-omics, and imaging. It outlines model development workflows, including preprocessing for class imbalance (e.g., SMOTE) and interpretability tools (e.g., SHAP), while addressing ethical and technical challenges.
RESULTS: Ensemble methods (e.g., Random Forest, XGBoost) and deep learning (e.g., CNNs) outperform logistic regression, achieving AUC values > 0.90. Key advancements include federated learning for privacy and bias mitigation strategies to enhance generalizability across populations.
CONCLUSIONS: Machine learning-based models enable predictive, preventive, and personalized obstetrics, facilitating early interventions and improved perinatal outcomes, though external validation and regulatory frameworks are essential for clinical adoption.
PMID:41817044 | DOI:10.1002/bdr2.70038