Abstract
OBJECTIVE: To develop and validate a predictive model for live birth following single vitrified-warmed blastocyst transfer (SVBT) by integrating early pregnancy ultrasound radiomics with clinical parameters. METHODS: This retrospective cohort study analyzed 925 SVBT cycles (2019-2022). Patients were randomly divided into a training set (n = 740) and a testing set (n = 185). Radiomics features were extracted from gestational sac and embryonic structures at 4 weeks post-transfer. Machine learning (ML) models were trained using Least Absolute Shrinkage and Selection Operator (LASSO) regression and validated with receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA). Model performance was compared among clinical, radiomics, and combined clinical-radiomics models. RESULTS: The combined clinical-radiomics model demonstrated the highest predictive performance (AUC = 0.806, training; 0.718, testing), outperforming the radiomics-only model (AUC = 0.786, training; 0.708, testing) and the clinical-only model (AUC = 0.673, training; 0.579, testing). DCA confirmed superior clinical utility, and calibration curves indicated excellent agreement between predicted and observed outcomes. CONCLUSION: Integrating ultrasound radiomics with clinical features significantly improves the prediction of live birth following SVBT. This model provides a novel, objective tool for personalized reproductive decision-making, improving embryo transfer strategies and patient counseling in assisted reproductive technology (ART).