Study Finds AI More Effective Than Sonographer Interpretation of Cardiac Function on Echocardiograms

Emerging research suggests that an artificial intelligence (AI) model is approximately 10 percent more accurate than sonographers at interpreting left ventricular ejection fraction (LVEF) on echocardiograms.

For the blinded, randomized trial, recently published in Nature, researchers compared initial sonographer assessment of echocardiograms for LEVF (1,755 studies) versus initial AI assessment (1,740 studies). According to the study, the AI model was trained with 144,184 echocardiogram videos from Stanford Healthcare. Twenty-five cardiac sonographers (with a mean of 14.1 years of experience in practice) reviewed the echocardiograms and 10 cardiologists (mean of 12.7 years of experience in practice) completed final reviews of the initial AI and sonographer assessments.

Cardiologists substantially changed 292 study assessments (16.8 percent) in the AI group and 478 interpretations (27.2 percent) in the sonographer group. Researchers also reported time savings for cardiologists reviewing AI findings, noting that cardiologists spent a median of 54 seconds to review initial AI assessment of LEVF on echocardiograms in comparison to a median of 64 seconds for review of sonographer findings.

“After blinded review of AI versus sonographer-guided LVEF assessment, cardiologists were less likely to substantially change the LVEF assessment for their final report with initial AI assessment. Furthermore, AI-guided assessment took less time for cardiologists to overread and was more consistent with cardiologist assessment from the previous clinical report,” wrote study co-author David Ouyang, M.D., who is affiliated with the Department of Cardiology at the Smidt Heart Institute and the Division of Artificial Intelligence in Medicine at the Cedars-Sinai Medical Center in Los Angeles, and colleagues.

(Editor’s note: For related content, see “Pie Medical Imaging Launches AI-Powered Echocardiography Platform” and “FDA Clears AI-Powered Echocardiography Platform for Detecting Heart Failure with Preserved Ejection Fraction.”)

The study authors also noted the majority of cardiologists could not tell the difference between initial AI and initial sonographer assessments with 43.4 percent being unsure and 24.2 percent guessing incorrectly about the source of the initial echocardiogram interpretation.

In addition to the inherent limitations of a single center study, the authors conceded their research was not powered for evaluation of long-term outcomes associated with differences in LVEF assessment. Noting that the study involved prospective evaluation of previously acquired echocardiogram studies, the researchers acknowledged a potential for bias when sonographers interpret images scanned by another sonographer.