Google Research says a prospective multi-center study found that its AI model could estimate gestational age from “blind sweep” ultrasound scans in two different clinical settings. The findings were published in JAMA Network Open under the title “Generalization of AI-Based Gestational Age Assessment Using Blind Sweep Ultrasonography.”

The blind sweep method involves gliding a portable ultrasound probe across the abdomen in a standardized pattern. The study assessed whether operators without specialized sonography training could collect scans that an AI system could interpret.

The model was trained on data from 1,968 participants in North Carolina and Zambia. Researchers fine-tuned it using scans from 120 participants in Chicago, then prospectively tested it on 192 participants at Northwestern Medicine in Chicago and 193 participants at Jacaranda Health in Nairobi. No Nairobi data was used for fine-tuning.

Among 385 participants between 16 and 36 weeks of pregnancy, the model had a combined mean absolute error of 4.18 days for gestational age estimates, compared with 4.48 days for expert sonographer biometry. The reported error was 4.10 days in Chicago and 4.27 days in Nairobi.

Google Research described the model as non-inferior to the expert sonographer approach in the study. The researchers also reported that the system performed well in an exploratory analysis involving pregnancies with fetal weight above the 90th percentile.

The researchers identified practical deployment considerations, including the need for swappable batteries and safeguards such as confidence estimates that can reject poor-quality scans and communicate uncertainty to clinicians.