From virtual pregnancy to digital twin obstetrics: multimodal data integration for personalized prediction of pregnancy complications.

Pregnancy complications, including preeclampsia, gestational diabetes mellitus, preterm birth, fetal growth restriction, and placental insufficiency, remain major contributors to maternal and neonatal morbidity worldwide. Conventional prediction models in obstetrics have traditionally relied on static clinical variables collected at discrete gestational time points. Although these approaches are clinically convenient, they are insufficient to capture the dynamic, heterogeneous, and nonlinear nature of pregnancy physiology. Recent advances in computational modeling, artificial intelligence, biomedical data acquisition, and continuous physiological monitoring have introduced the concepts of virtual pregnancy and digital twin obstetrics. Virtual pregnancy modelsaim to simulate maternal-placental-fetal interactions through mechanistic or computational frameworks, whereas digital twins aim to extend this concept by integrating real-time multimodal data streams to construct continuously updated, patient-specific models. This review discusses the conceptual transition from virtual pregnancy to digital twin obstetrics and examines how multimodal data integration can support personalized prediction of pregnancy complications. Particular attention is given to clinical phenotyping, obstetric imaging, radiomics, multi-omics data, wearable-derived physiological monitoring, and machine learning-based modeling strategies. The review further evaluates potential applications in preeclampsia, gestational diabetes mellitus, preterm birth, fetal growth restriction, and placental dysfunction. Although digital twin obstetrics represents a promising but still largely conceptual framework for precision maternal-fetal medicine, major challenges remain, including data heterogeneity, limited external validation, model interpretability, ethical governance, privacy protection, and clinical implementation. Future work should prioritize standardized data infrastructures, prospective multicenter validation, explainable artificial intelligence, and clinically interpretable decision-support systems.