Real-time gestational-age estimation at 11-13 weeks using artificial intelligence on non-targeted ultrasound sweeps.
Authors
Affiliations (3)
Affiliations (3)
- Fetal Medicine Research Institute, King's College Hospital, London, UK.
- Institute of Health Research, University of Exeter, Exeter, UK.
- Department of Women & Children's Health, School of Life Course & Population Sciences, Faculty of Life Sciences & Medicine, King's College London, London, UK.
Abstract
To assess the accuracy of gestational-age estimation using an artificial intelligence (AI) system applied to non-targeted ultrasound sweeps at 11 + 0 to 13 + 6 weeks' gestation. This was an observational study of singleton pregnancies undergoing routine first-trimester ultrasound in July 2025 to August 2025. Gestational age was estimated using crown-rump length (CRL), measured independently by two experienced sonographers blinded to the other's results, and by an AI system applied twice per examination. Comparisons were made using mean absolute error (MAE) and Bland-Altman analysis. A hierarchical measurement error model was used to account for uncertainty in CRL-based estimates and to estimate accuracy relative to a latent true gestational age, with Bayesian inference. Overall, 285 women with a singleton pregnancy were included. The AI model estimated gestational age in all cases, with a median acquisition time of 23 s for the first AI measurement. The MAE between AI- and CRL-based gestational age was 2.1 (95% CI, 1.9-2.3) days for single measurements and 2.0 days when averaging both AI estimates. Limits of agreement were wider for AI-based estimates than for CRL-based estimates, indicating lower precision. When accounting for measurement error in CRL, the estimated MAE for AI was lower (1.8 days for single and 1.6 days for duplicate measurements). CRL-based estimates showed lower variability than AI-based estimates. Within-operator variation was the main source of error in AI measurements. AI-based gestational-age estimation performs well in the first trimester but is less precise than gestational-age estimation using CRL measured by experienced sonographers. Measurement error models give a clearer picture of performance by accounting for variability in CRL and show that the accuracy of AI-based estimates is better than suggested by simple comparisons. © 2026 The Author(s). Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.