AI Ultrasound Matches Experts Within 4.18 Days Using 385 Pregnancies and Low-Cost Probes
Updated
Updated · Google Research · Oct 21
AI Ultrasound Matches Experts Within 4.18 Days Using 385 Pregnancies and Low-Cost Probes
2 articles · Updated · Google Research · Oct 21
Summary
A prospective study in JAMA Network Open found an AI system could estimate gestational age from blind-sweep ultrasounds with a 4.18-day mean absolute error, versus 4.48 days for expert sonographers.
The model was designed for minimally trained, non-sonographer operators using portable low-cost probes, aiming to expand prenatal imaging where specialist training and standard ultrasound access are scarce.
Across 385 pregnancies tested prospectively in Chicago and Nairobi, accuracy held steady at 4.10 and 4.27 days, even though the system received no Nairobi fine-tuning.
Training used 1,968 participants from North Carolina and Zambia, then just 120 Chicago cases—about 6% of the original training set—to adapt to new hardware and extend predictions beyond 36 weeks.
The findings matter because two-thirds of people globally lack access to diagnostic imaging, and WHO estimates pregnancy-related complications still cause more than 250,000 preventable deaths each year.
Could relying on AI-guided blind sweeps cause minimally trained workers to miss critical fetal anomalies that experts would easily spot?
How will rural clinics manage the hidden costs and electronic waste of maintaining battery-powered AI ultrasound probes in harsh environments?
From Novice to Expert: AI Handheld Ultrasound Delivers Sonographer-Equivalent Gestational Age Estimation (MAE 3.2 Days) in 1.8 Million Global Scans
Overview
Recent clinical breakthroughs have shown that AI-powered handheld ultrasound devices allow novice healthcare workers to estimate gestational age with accuracy matching expert sonographers, even in low-resource settings. By replacing manual image capture with simple blind sweeps and running AI algorithms directly on mobile devices, these systems deliver real-time results without internet access. Streamlined training enables frontline workers to become proficient in just one day, while large-scale deployments have integrated thousands of devices across hundreds of clinics. This approach not only boosts provider confidence and reduces scan errors but also makes expert-level diagnostics affordable and accessible worldwide, driving global health equity.