Artificial intelligence (AI) may reduce cognitive load for sonographers in fetal ultrasound scans without affecting diagnostic performance
1. Day and colleagues randomized sonographers to scan pregnant participants with AI assistance or the standard method for congenital heart disease screening.
2. AI-assisted scans demonstrated similar sensitivity and specificity to standard scans, while being significantly shorter in duration.
Evidence Rating Level: 1 (Excellent)
Study Rundown: Fetal anomaly ultrasound (US) screening is a common method used to diagnose congenital malformations antenatally. However, universal detection of major fetal malformations, such as congenital heart disease (CHD), has not been achieved. AI assistance has the potential to improve fetal anomaly screening effectiveness. Day and colleagues conducted a randomized controlled trial that compared AI-assisted fetal US scans to the standard technique. Sonographers were randomly assigned to perform scans either with or without AI assistance on pregnant patients with healthy fetuses or fetuses diagnosed with CHD. The primary outcomes were the sensitivity and specificity of the two methods in detecting CHD, and secondary outcomes were the time taken to complete the US scan and report and the cognitive load of the sonographers. The study found that there were no statistically significant differences between the two methods for detecting fetal CHD or all fetal structural malformations. However, on average, the AI-assisted group took 9.4 minutes less than the standard group to complete the US scan and report. This study demonstrated AI’s ability to improve fetal US screening efficiency without compromising diagnostic performance.
Click here to read the study in NEJM AI
Relevant Reading: Recent Advances in Artificial Intelligence-Assisted Ultrasound Scanning
In-Depth [randomized controlled trial]: Fifty-nine sonographers were randomized into the AI-assisted (intervention) and standard (control) groups in a 1:1 ratio. All sonographers underwent a standardized training session, followed a study-specific protocol, and were blinded to the clinical status of the pregnant participants. On a given day of the trial, three pregnant participants (two with a healthy fetus and one with a fetus with CHD) were scanned by one sonographer from each group. The primary outcomes were the sensitivity and specificity of the two methods in detecting CHD, and secondary outcomes were the time taken to complete the US scan and report and the cognitive load of the sonographers measured by a survey (NASA-TLX). For detecting fetal CHD, the difference in sensitivity and specificity between the two groups was 3.8% (97.5% confidence interval (CI), -18.9-26.6%) and 7.7% (97.5% CI, -0.6-16.0%), indicating statistical insignificance. Meanwhile, the median scan duration was shorter for the intervention group by 9.3 minutes (95% CI, 7.4-11.1). The sonographers in the intervention group also experienced reduced cognitive load than the control group (mean difference in NASA-TLX score: 10.0, 95% CI, 4.6-15.4). The limitations of this study included the self-selection of sonographers (who may not represent the broader workforce), the knowledge of a malformation being present despite being blinded to the specific diagnosis of each participant, and increased caution in a small research setting. Nonetheless, this study provided encouraging evidence that AI-assisted fetal US screening effectively reduced scan duration and cognitive load without reducing diagnostic performance.
Image: PD
©2025 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.
