Google reports AI ultrasound study matched expert pregnancy dating accuracy
Google Research reported that AI analysis of ultrasound scans collected by novice operators was non-inferior to expert sonographer assessments in a prospective study in Chicago and Nairobi. The findings concern gestational age estimation; wider access remains a potential benefit.
Google Research said on 2026-10-21 that a prospective study in Chicago and Nairobi found its research AI system could estimate gestational age from standardized ultrasound “blind sweeps” collected by operators without specialized sonography training, with accuracy non-inferior to expert sonographer assessments.
According to Google's account of the study published in JAMA Network Open, the primary cohort included 385 participants between 16 and 36 weeks of pregnancy. Mean absolute error was 4.18 days for the AI system and 4.48 days for the expert approach.
The model was fine-tuned using scans from 120 Chicago participants. No Nairobi data was used for fine-tuning, making that site's evaluation an independent external test, Google said.
Google described broader prenatal ultrasound access as a potential application. The reported findings assess pregnancy dating accuracy; Google also identified safeguards that convey uncertainty and reject poor-quality scans as important for future deployment.