Google researchers are testing a way to use artificial intelligence to expand access to prenatal ultrasound scans, particularly in areas that lack specialized equipment or medical imaging technicians. The approach involves training non-specialist health workers to perform a simplified abdominal scan, then analyzing the recordings with a machine-learning model running on the device itself.
The study was conducted in collaboration with Jacaranda Health and Northwestern Medicine and included 1,000 mothers in Nairobi, Kenya, and another 1,000 mothers in Chicago. According to the researchers, the model was able to determine gestational age and fetal position with accuracy comparable to that of a trained ultrasound technician.
How does the idea work?
In a traditional examination, the technician must move the probe precisely to obtain specific measurements, a skill that requires extensive hands-on training. The approach the study calls a “blind sweep” instead involves passing the probe over the abdomen according to a predefined pattern. The researchers say that a health worker with no prior experience performing ultrasound scans can learn the procedure within eight hours.
The health worker records a video during the scan, which is then analyzed by the artificial intelligence model. The model can alert the operator when the scan needs to be repeated, estimate gestational age, and determine the fetus’s position, meaning its orientation inside the uterus.
What changes in practice?
The importance of the approach lies in the fact that it does not attempt to replace all the functions of a medical imaging technician, but instead focuses on basic information that could affect preparation for delivery. Accurately determining gestational age helps doctors interpret fetal growth, estimate the due date, and prepare to handle cases requiring preterm delivery. The source explained that an error in estimating gestational age can lead to a significant difference in the level of care required for newborns, particularly when a 37-week fetus is confused with one that is 34 weeks old.
Processing also takes place on the device, without requiring a Wi-Fi connection or a continuous power supply. This feature addresses some of the constraints faced by remote clinics, where traditional equipment is large and expensive and may be difficult to repair when it breaks down, while portable devices offer a smaller, lower-cost, battery-powered option.
Limits of the result and what the study did not establish
The results show the possibility of expanding access to two specific indicators—gestational age and fetal position—but do not establish that the system can replace comprehensive medical assessment or specialized technicians for all types of pregnancy scans. The source also does not provide details about regulatory approval, detailed error rates, or a commercial availability plan, so the result remains within a research framework demonstrating the feasibility of the idea.
For certi.news, the actual change here is not the launch of a new device, but the transfer of part of the ultrasound imaging skill to a workflow that can be learned within hours, followed by the use of a locally processed model to extract specific clinical information. Its impact will still depend on the quality of the devices, training and medical supervision, and the ability of health systems to act on the results.