Maternal–Fetal Oncology Intelligence: Self-Evolving Foundation Models and Digital Twin Ecosystems for Precision Cancer Prediction, Treatment Optimization, and Maternal–Fetal Outcome Forecasting
Year: 2026
Type: article
Abstract: Background.Cancer during pregnancy poses a dual clinical challenge: maternal disease must be controlled without compromising fetal safety, yet prediction of outcomes is hampered by fragmented multimodal data and the absence of models that adapt as pregnancy advances.We developed a maternal-fetal oncology framework combining a multimodal foundation model with continuously updated patient-specific digital twins to support cancer prediction, risk stratification, treatment optimization, and joint maternal-fetal outcome forecasting. Methods.A retrospective cohort of 350 pregnant women diagnosed with cancer was assembled from four tertiary centers, spanning breast, cervical, ovarian, thyroid, melanoma, lymphoma, and leukemia.The framework (MFO-FM) integrates clinical, imaging, genomic, and laboratory encoders through a maternal-fetal fusion module, while a digital twin ecosystem assimilates longitudinal maternal physiology, fetal physiology, cancer progression, and treatment response.Performance was benchmarked against logistic regression, random forest, XGBoost, and a deep neural network using five-fold cross-validation and an external test set.Results.MFO-FM achieved an accuracy of 0.927, F1-score of 0.916, and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.958, exceeding XGBoost (AUC 0.911) and the deep neural network (AUC 0.928).Favorable maternal outcomes occurred in 262 of 350 women (74.9%) and healthy neonatal outcomes in 214 of 350 (61.1%), with preterm birth in 97 (27.7%).Cancer stage, gestational age at diagnosis, and chemotherapy timing were the dominant predictors.The digital twin reproduced observed trajectories with a concordance index of 0.85.Conclusions.An adaptive foundation model coupled with digital twin ecosystems delivered accurate, interpretable, trimester-aware predictions across the maternal-fetal oncology continuum.Prospective multicenter validation is required before clinical adoption.
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W7221553779