Randomised trial of AI feedback on clinicians reading fetal Doppler ultrasound images
96 clinicians at five Danish hospitals were randomised to receive AI feedback or none while judging 40 standardised umbilical and middle cerebral artery Doppler vignettes each, and those given feedback were accurate on 75.0% of vignettes against 62.3% without it, a difference of 12.7 percentage points measured against expert consensus rather than patient outcomes.
- Model
- An explainable AI system giving visual and textual feedback on Doppler gate placement, anatomical suitability, insonation angle and waveform quality (not named)
- Field
- Medicine
- Date
- 2026-11-06
- Human collaborators
- Zahra Bashir, Aasa Feragen, Martin G. Tolsgaard
What was found
Umbilical and middle cerebral artery Doppler measurements guide management of fetal growth restriction, and their reliability depends heavily on who is holding the probe. Residents, gynaecologists and obstetricians were randomised 1:1 and matched into 48 pairs by obstetric ultrasound experience, each evaluating 40 vignettes built from pre-existing static Doppler images, 1,920 unique vignettes in all. For colour Doppler images clinicians indicated where they would place the measurement gate, and for spectral images they judged whether the waveform was technically adequate, both components of a valid measurement. Intervention clinicians saw the unannotated image alongside the AI's feedback, then entered their own answer on the unannotated image. Accuracy was scored against consensus between two fetal medicine sonographers with more than ten years of experience, with a consultant resolving disagreements, using criteria based on International Society of Ultrasound in Obstetrics and Gynecology guidance. The authors report that clinical experience still mattered when the AI feedback was itself wrong, and frame the result as support for studying human and AI collaboration during real-time acquisition.
Novelty check
Europe PMC, 2026-11-09: TITLE:(Doppler) AND ("artificial intelligence" OR "deep learning" OR "AI") AND (umbilical OR "middle cerebral") AND (randomized OR randomised) returned 24 records, of which this is the only randomised trial of AI feedback on fetal Doppler assessment; the nearest prior work is a 2024 Frontiers in Digital Health paper on an automated classification workflow for feto-placental Doppler images, with no clinician randomisation, and a 2026 meta-analysis of interobserver variability in these measurements that establishes the problem being addressed. No entry in this registry covers obstetric Doppler.
Caveats and known objections
This sits at the edge of what this registry counts as a clinical result, and is included because clinicians judged real anonymised case material with their performance measured, not an exam: no scanning was performed, no patient was managed, and static vignettes remove the probe handling that makes live Doppler hard, so the 12.7 point gain should not be read as an effect on fetal care. The reference standard is expert consensus, which is itself fallible on exactly the measurements the meta-analysis above shows experts disagree about. Clinicians in the intervention arm saw the AI's answer before giving their own, so part of the measured gain is agreement with the AI where the AI agreed with the experts, and the authors note that experience mattered when the feedback was wrong. One country, five hospitals, 96 clinicians, a single specialty. The system is not named, no version is given and nothing about it is available to outside investigators. The abstract cites no trial registration; a registration matching this design, sponsor and endpoint appears at NCT06371859, which lists 92 clinicians and a status of unknown, and the paper does not confirm it, so it is recorded here rather than as a registration. Autonomy is ai-assisted: the clinician entered every answer and the system commented.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Peer reviewed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded peer reviewed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Randomised trial of AI feedback on clinicians reading fetal Doppler ultrasound images. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-11-06-doppler-ai-feedback-rct
BibTeX
@misc{whataifound-universityofcopenhagentechnicaluniversityofdenmarkandcopenhagenacademyformedicaleducationandsimulation-2026-rct,
title = {Randomised trial of AI feedback on clinicians reading fetal Doppler ultrasound images},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by University of Copenhagen, Technical University of Denmark and Copenhagen Academy for Medical Education and Simulation. Verification: Peer reviewed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-11-06-doppler-ai-feedback-rct}
}