Google Says AI Matches Sonographers on Two Prenatal Ultrasound Measures

Google researchers want to put expert-level prenatal information within reach of clinics short on ultrasound specialists. In an October 6 interview published by Google, they described a narrower but concrete result: AI interpreted simple ultrasound videos as accurately as a trained sonographer on two measures—how far along a pregnancy is and the baby’s position.

The study, conducted with Jacaranda Health and Northwestern Medicine, involved 1,000 mothers in Nairobi, Kenya, and another 1,000 in Chicago. Google AI researcher Angelica Willis described the findings alongside Dr. Nichole Young-Lin, the company’s in-house obstetrician-gynecologist. Their account presents the work as research into widening access to maternal care.

A simpler route to the scan

A conventional ultrasound requires a sonographer to guide the probe precisely to obtain specific measurements. The study instead used a “blind sweep”: a healthcare worker moves the probe across the abdomen in a predefined pattern, collecting videos for the AI to analyze.

Willis said healthcare workers who had never performed an ultrasound could learn this procedure in eight hours. That is training for the sweep, not for the full work of a sonographer. She contrasted it with the two years of intensive hands-on training needed to become an ultrasound specialist.

The model can tell the operator to repeat a sweep. It also estimates gestational age—the stage of pregnancy—and indicates fetal presentation, meaning the baby’s position. Willis said the processing happens on the device without Wi-Fi, an important design detail for the remote clinics the researchers hope to reach.

Why pregnancy dating carries weight

Young-Lin explained that gestational age sets the timeline for screening, treatment and planning a safe delivery. Menstrual history is not always a reliable substitute: some patients have irregular periods or do not track them. Early ultrasound helps clinicians estimate when a baby is due.

She illustrated the stakes with a hypothetical early delivery: a baby believed to be at 37 weeks might actually be at 34 weeks, with less mature lungs. Those newborns need different levels of care. Accurate dating helps teams prepare support for both the mother and baby.

The access problem is not just equipment. Young-Lin said handheld scanners are smaller, cheaper and often battery-powered, while conventional machines can be difficult to repair in remote clinics. Willis pointed to specialist shortages as another barrier. The study’s reported accuracy addresses two specific measurements; Google’s interview gives no numerical error rates for that comparison.