Summary
Google describes a study with 2,000 mothers in Nairobi and Chicago that used AI to interpret video from simple “blind sweep” ultrasound scans. The company says the model estimated gestational age and fetal presentation as accurately as a trained sonographer.
Google researchers have tested an AI-assisted approach to prenatal ultrasound in a study involving 1,000 mothers in Nairobi, Kenya, and 1,000 in Chicago. In an account published by Google, the team says the model estimated gestational age and fetal presentation as accurately as a trained sonographer. The work was conducted with Jacaranda Health and Northwestern Medicine.
The approach is designed around a simpler scan that frontline health workers can learn to perform without becoming specialist sonographers. It could make ultrasound services easier to provide in settings where trained operators and conventional equipment are scarce, although the report describes a research study rather than a measured expansion of clinical access.
A scan built around a guided sweep
In a conventional ultrasound, an operator moves the probe to obtain particular views and measurements. The researchers instead used a “blind sweep”: the operator moves the probe over the abdomen in a predefined pattern, recording video. Google’s researchers say health workers with no previous ultrasound experience can learn the procedure in eight hours.
The AI model analyses the video, can indicate whether the operator should repeat the sweep, and estimates gestational age—the stage of pregnancy—and fetal presentation, or the baby’s position. Google says the processing happens on the device and does not require Wi-Fi. The post does not describe the particular device used.
Why those estimates matter
Gestational age helps clinicians plan screening, treatment and delivery care. As the Google interview explains, a pregnancy’s stage can affect the level of support a newborn may need, particularly if delivery occurs early. Fetal presentation also provides information about the baby’s position ahead of birth.
Portable ultrasound devices may ease equipment barriers, but operating a scan has traditionally required specialist training. A workflow that combines a repeatable sweep with automated interpretation could allow more healthcare workers to collect useful images. Google presents the study as a step towards broader access, not as evidence that the approach has already changed how many people receive prenatal scans.
The company’s account summarizes the accuracy comparison but gives no performance figures or detailed study methods. That makes the reported finding difficult to assess beyond Google’s description; the post does not establish how the model would perform across different clinics or in routine care.