In a blinded study, artificial intelligence (AI) performed better than trained sonographers in reading echocardiograms of the heart.
The research, led by Smidt Heart Institute and the Division of Artificial Intelligence in Medicine at Cedars-Sinai, showed there was less disagreement between the first assessment made by AI and a final cardiologist assessment than there was if the first scan was assessed by a sonographer.
“The results have immediate implications for patients undergoing cardiac function imaging as well as broader implications for the field of cardiac imaging,” said cardiologist David Ouyang, principal investigator of the clinical trial and senior author of the Nature paper describing the work.
“This trial offers rigorous evidence that utilizing AI in this novel way can improve the quality and effectiveness of echocardiogram imaging for many patients.”
Work by the same group, published in 2020, developed AI technology to assess left ventricular ejection fraction, a measure commonly used to evaluate cardiac function. However, this technology was not previously tested in a randomized fashion.
To test the effectiveness of the AI developed by Ouyang and team, a blinded, randomized study was carried out comparing the accuracy of the AI assessment of patient electrocardiograms with that of trained sonographers.
The primary end point of the study was the change in assessment of left ventricular ejection fraction between the first evaluation, carried out by the AI or a sonographer, and the final expert cardiologist assessment. This was measured as the proportion of cases with a change greater than five percent.
Overall, 3495 electrocardiograms were included in the analysis. The trial met the primary endpoint. The percentage of scans with a substantial change in diagnosis was 16.8% in the AI group vs 27.2% in the sonographer group, with an average absolute difference of 6.29% for the AI group and 7.23% for the sonographer group.
“Cardiologists required less time, substantially changed the initial assessment less frequently and were more consistent with previous clinical assessments by the cardiologist when using an AI-guided workflow. This finding was consistent across subgroups of different demographic and imaging characteristics,” write the authors.
“In the context of an ongoing need for precision phenotyping, our trial results suggest that AI tools can improve efficacy as well as efficiency in assessing cardiac function.”
The researchers now plan to roll-out use of the AI tool to multiple medical centers and continue to monitor its diagnostic performance against more traditional methods.