Multimodal AI-Augmented Radiomics and Spectral Doppler Ultrasound for Early Prediction of Pre-Eclampsia

AbstractPre-eclampsia remains a major cause of global maternal morbidity and mortality. Existing predictive models have limitations in sensitivity, specificity, and biomarker standardization. This study evaluates AI-driven ultrasound biomarkers that integrate self-supervised deep learning (SSDL), hybrid graph neural networks (HGNNs), and Bayesian hierarchical modeling to enhance early risk prediction. The goal is to improve stratification using placental texture analysis, hemodynamic modeling, and maternal serum biomarkers. A systematic review and meta-analysis following PRISMA guidelines will synthesize data from PubMed, Embase, Scopus, IEEE Xplore, and the Cochrane Library. Eligible studies must assess AI-enhanced ultrasound biomarkers for pre-eclampsia prediction. AI models incorporate federated learning for generalizability and explainability frameworks (SHAP, Grad-CAM, LIME) for transparent clinical interpretation. Comparative analyses are conducted against traditional classifiers, and a bivariate random-effects meta-analysis estimates pooled diagnostic accuracy metrics. Preliminary findings suggest that AI-enhanced radiomics and spectral Doppler biomarkers achieve AUC-ROC values between 0.90–0.97, outperforming conventional methods. Multi-center analyses show statistical superiority (p < 0.001) over standard screening approaches, confirming cross-populational reproducibility. These findings demonstrate that AI-enhanced placental imaging and hemodynamic modeling enable earlier risk stratification, improved decision-making, and reduced diagnostic uncertainty. ConclusionsThis study supports AI-augmented radiomics and spectral Doppler ultrasound as a scalable, interpretable, and globally adaptable framework for pre-eclampsia prediction. The combination of Bayesian probabilistic modeling and radiomics-based profiling addresses traditional sonographic limitations while explainability frameworks enhance physician trust in AI-assisted maternal–fetal diagnostics. Conference:35th World Congress on Ultrasound in Obstetrics and Gynecology (ISUOG 2025), Cancún, Mexico, 14–17 September 2025 Originally Published in:Ultrasound in Obstetrics & Gynecology (2025), Vol. 66 (Suppl. 1): EP16.66DOI: 10.1002/uog.70091 Zenodo DOI (Author Version):10.5281/zenodo.17333142 Author:Sunday A. Adetunji, MDDepartment of Epidemiology, College of Health, Oregon State University, USADepartment of Medicine and Surgery, Obafemi Awolowo University, Nigeriahttps://orcid.org/0000-0001-9321-9957 Funding: None declaredConflict of Interest: The author declares no conflicts of interest.License: Creative Commons Attribution 4.0 International (CC BY 4.0)Keywords: pre-eclampsia, ultrasound, spectral Doppler, radiomics, AI, maternal health, deep learning, Bayesian modeling, hybrid graph neural networks, ISUOG, Oregon State University