From the source
Google Research published a prospective multi-center study in JAMA Network Open showing that an AI system reading blind sweep ultrasounds matched expert sonographers at estimating gestational age, using novice operators and low-cost handheld probes.
- The World Health Organization estimates more than 250,000 preventable deaths occur worldwide each year from pregnancy-related complications, and two-thirds of people globally lack access to diagnostic imaging such as ultrasound.
- Blind sweep scanning lets frontline health workers without sonography training glide portable probes across the abdomen in a standardized pattern, an approach Google says could widen prenatal screening where experts are scarce and training is expensive.
- The model was trained on 1,968 participants in North Carolina and Zambia, fine-tuned on 120 participants from Chicago, then tested prospectively in Chicago (n=192) and Nairobi (n=193), with no Nairobi data used for tuning.
- Across a primary cohort of 385 participants between 16 and 36 weeks of pregnancy, the AI system reached a combined mean absolute error of 4.18 days versus 4.48 days for expert sonographers, which the researchers describe as non-inferior.
- The system ran live on mobile devices with the Clarius C3 handheld probe, and the study also included multiple gestation pregnancies.