BIOBRIEFPAPER
A Lancet Digital Health RCT in five Chinese centres found PAICS, a prenatal ultrasound AI, raised detection of specific fetal brain malformations by 8.7 to 11.8 points without raising false alarms.
A multicentre, self-crossover randomized trial in China tested whether an AI "second reader" could make prenatal ultrasound better at catching fetal brain malformations without raising false alarms.
The trial enrolled sonographers at five Chinese centres who scanned 1,584 high-risk singleton pregnancies at 11–32 weeks' gestation between September 2022 and November 2023. Each sonographer read cases twice, once independently and once with PAICS (Prenatal Ultrasound Diagnosis Artificial Intelligence Conduct System), in randomized order with a four-week washout. Allocation was concealed; outcomes assessors and the expert-panel reference diagnosis were masked to the AI arm. Sonographers themselves were not blinded to AI help during scanning.
Against the expert panel, sensitivity for ten specific fetal brain malformations rose by 0.087 (95% CI 0.029–0.147; p<0.0001) on a fetal-based analysis and by 0.118 (95% CI 0.054–0.181; p<0.0001) on a malformation-targeted analysis. Specificity difference was 0.009 (95% CI −0.006 to 0.023), clearing a pre-set 5% non-inferiority margin.
The result, published in Lancet Digital Health, comes with limits: a single country, sonographers with 3 to <8 years' experience, AI help disclosed in real time, and findings restricted to malformations within the system's recognition scope. Beyond-scope anomaly detection and any effect on counselling or delivery decisions are outside this readout.
What is not yet shown: whether the sensitivity gain holds in junior trainees, community settings, or low-risk pregnancies, and how the AI performs when sonographers do not know it is watching.