AI drafts cut heart-scan reporting time by 46% in preliminary trial

Cardiologists completed heart-ultrasound reports faster with AI-written drafts in a randomized study presented November 9 at the American Heart Association’s Scientific Sessions 2026. The drafts met the study’s standard for performing no worse than human-prepared reports. But doctors still substantially revised about half of both groups, making this a test of assistance—not autonomous diagnosis.

The findings are preliminary. They were presented as a research abstract, which the association says has not been peer-reviewed.

Three first drafts, one set of scans

An echocardiogram is an ultrasound scan of the heart. Usually, a sonographer acquires the images and a cardiologist reviews them and writes a report. The study tested AI after images had already been collected.

Researchers used 1,200 routine-care exams performed over two weeks in March 2025, across 21 Kaiser Permanente Northern California medical centers. Nine sonographers and 20 cardiologists participated.

For each scan, an AI tool, a sonographer and a cardiologist each prepared preliminary reports. Those drafts were randomly assigned to cardiologists for review and finalization. Reviewers did not know who—or what—had prepared the first version.

The researchers describe it as the first randomized trial of an AI tool drafting a full echocardiogram report, including measurements and interpretations—not just an isolated measurement.

Similar revision rates, not untouched reports

The study assessed accuracy by counting how often a cardiologist made substantial changes before signing off. That happened to 50.6% of AI-generated interpretations, compared with 49.7% of human-prepared drafts. The AI approach met the study’s noninferiority criteria—its threshold for being considered comparable to human-assisted reporting.

Those percentages describe edits to preliminary interpretations, not the proportion of patients correctly diagnosed. A cardiologist completed the report in every group.

AI-assisted reports were also more consistent with historical reports in patients’ medical records. Reviewing cardiologists could not identify whether a draft came from AI, a sonographer or another cardiologist.

Minutes saved versus care improved

Co-senior author Keane K. Lee, a Kaiser Permanente cardiologist, said accumulated time savings could potentially mean faster results and greater access to care. The study did not test whether AI assistance improves patients’ long-term health. Lee emphasized that cardiologists remain responsible for certifying final reports and applying clinical judgment to care.

Cardiologists knew they were participating in a study and may have changed their usual work patterns. The authors called for larger-scale research with higher exam volumes to assess performance in real-world clinical settings.