An AI system using simple handheld ultrasound sweeps matched the pregnancy-dating accuracy of expert sonographers in a prospective study spanning clinics in Chicago and Nairobi.
Frontline workers without specialist sonography training moved a low-cost probe across the abdomen in a standardized pattern. The model estimated gestational age with a combined mean absolute error of 4.18 days, compared with 4.48 days for conventional measurements performed by experts. The primary cohort included 385 pregnancies between 16 and 36 weeks.
Researchers trained the original model on 1,968 participants in North Carolina and Zambia, then adapted it with scans from 120 Chicago participants to support new hardware and later pregnancies. It achieved 4.27-day error in Nairobi without any local fine-tuning, providing an external test across different patients and workflows.
The result demonstrates technical generalization, not a finished replacement for prenatal care. Researchers say deployment still needs confidence estimates that reject poor scans, clinician trust and practical provisions such as reliable battery power. The system estimates fetal age; it does not perform every diagnostic task of a full ultrasound examination.