AI Tool Matches Sonographers in Prenatal Ultrasound Dating

AI achieves clinical-grade accuracy in gestational age assessment

A portable artificial intelligence system can estimate the gestational age of unborn babies as accurately as trained sonographers, according to research that could dramatically expand access to prenatal care in underserved regions.

The AI model, developed by researchers at Google, achieved a mean absolute error of 4.2 days when estimating gestational age from ultrasound scans—performance that met the clinical standard of care. The system worked effectively even when operated by novices using low-cost portable ultrasound probes across geographically and economically diverse settings.

Researchers Ryan Gomes and colleagues published their findings in JAMA Network Open, demonstrating the tool's performance across 2,043 participants in Chicago and Nairobi. The study represents a significant step toward making essential prenatal diagnostics accessible where skilled sonographers and expensive equipment remain scarce.

How the system works across diverse settings

The AI model relies on blind sweep ultrasonography—a standardized scanning protocol that doesn't require real-time image interpretation by the operator. This approach makes it feasible for minimally trained users to capture diagnostic-quality scans.

Initially trained on data from suburban North Carolina and urban Zambia, the system required minimal fine-tuning to adapt to new environments. Using just 180 examinations from 120 Chicago participants—roughly 6 percent of the original training dataset—researchers successfully adapted the model to work with different hardware and patient populations.

In the primary evaluation involving 385 participants with gestational ages between 16 and 36 weeks, the system performed consistently across both test sites. In Chicago, the mean absolute error was 4.1 days, while in Nairobi it reached 4.3 days without any local tuning—demonstrating robust transfer learning capabilities.

Why it matters

Accurate gestational age determination is fundamental to prenatal care, enabling timely interventions that prevent complications from pre-term and post-term births. The World Health Organization recommends ultrasonography for all pregnant women, yet many low-resource settings lack both the skilled personnel and expensive equipment required for traditional ultrasound examinations. An AI system that operates reliably with minimal training and affordable hardware could address a critical gap in maternal healthcare infrastructure, potentially reducing preventable maternal and fetal complications in underserved populations worldwide.

Training requirements and practical implementation

The research revealed an interesting operational insight: operators in Nairobi who received approximately six hours of formal, interactive, hands-on training achieved a lower sweep rejection rate (1.8 percent) compared to Chicago operators who received informal written instruction (7.9 percent rejection rate). This suggests that even modest structured training can significantly improve scan quality.

The researchers noted that the system's generalizable accuracy with low-cost probes represents an important step toward scalable prenatal implementation aligned with WHO recommendations. They suggest the approach could serve as a template for automating other ultrasound-based diagnostics in obstetric care.

The findings were first reported by Inside Precision Medicine, based on the research published in JAMA Network Open.