Kaiser Permanente investigator’s validation study sets stage to bring EchoPrime into clinical care
An international team of investigators has built the largest-ever artificial intelligence (AI) model to produce interpretations of echocardiograms — cardiac ultrasound tests that health care providers use to diagnose and treat heart attacks, valve disease, heart failure, or other heart problems.
“We built this AI model to help clinicians quickly and easily provide more accurate and precise interpretations of echocardiograms to patients,” said co-senior author David Ouyang, MD, a research scientist with the Kaiser Permanente Division of Research and a non-invasive cardiologist and echocardiographer with The Permanente Medical Group. “Echocardiograms are the most commonly used cardiac imaging test, and the AI model we developed will benefit researchers, clinicians, and, most importantly, patients.”
The findings, published in Nature, highlight the comprehensive reporting capabilities of the AI model, called EchoPrime.
Existing AI models are trained on limited datasets and often only assess one or two specific heart functions. EchoPrime is a new vision-language AI model with the ability to learn from both images and text. It was trained on more than 12 million videos paired with expert cardiologists’ interpretations.
We built this AI model to help clinicians quickly and easily provide more accurate and precise interpretations of echocardiograms to patients.
— David Ouyang, MD
“The scale of training data is very important in developing accurate and reliable AI models,” explained co-senior author Bryan He, PhD, a research scientist at Stanford University. “EchoPrime is trained on over 1,000 times more data than our prior work, and more than 10 times more data than the largest existing AI models for echocardiography.”
EchoPrime can simultaneously interpret a wide range of cardiac features and diagnoses and detect hundreds of conditions that cardiologists frequently assess when interpreting an echocardiogram.
“Cardiologists analyze and integrate information from multiple views and videos to make a diagnosis,” said Ouyang. “EchoPrime was designed to evaluate complementary information of multiple videos in the same way and to then create a comprehensive written report of the most important aspects of the heart’s form and function.”
The research team tested and validated the AI model on clinical echocardiography interpretation tasks at 5 international health care sites: Kaiser Permanente Northern California, Cedars-Sinai Medical Center, Stanford Healthcare, Beth Israel Deaconess Medical Center, and Chang Gung Memorial Hospital in Taiwan. The study showed that EchoPrime outperformed all existing echocardiography and general purpose AI models in echocardiography interpretation tasks.
A clinical trial of EchoPrime use in preliminary assessment of echocardiography is now underway at Kaiser Permanente Northern California.
The EchoPrime model is open source and can be used by researchers to study AI for echocardiography.
“EchoPrime is not only the largest existing echocardiography AI model, but we are publicly releasing the model so that it can be used to advance research in AI in medicine worldwide,” said He. “We believe it has the potential to improve the reliability, accuracy, and cost-effectiveness of echocardiography, allowing more patients to benefit from this critical cardiac test.”
The study was funded by the National Heart, Lung, and Blood Institute.
Co-authors include Milos Vukadinovic of the University of California, Los Angeles; Debiao Li, PhD, and Susan Cheng, MD, of Cedars-Sinai Medical Center; I-Min Chiu, MD, PhD, and Tien-Yu Chen, MD, of Kaohsiung Chang Gung Memorial Hospital; Xiu Tang, RCDS, and Paul Cheng, MD, PhD, of Stanford University; and Neal Yuan, MD, of the University of California San Francisco.
###
About the Kaiser Permanente Division of Research
The Kaiser Permanente Division of Research conducts, publishes, and disseminates epidemiologic and health services research to improve the health and medical care of Kaiser Permanente members and society at large. KPDOR seeks to understand the determinants of illness and well-being and to improve the quality and cost-effectiveness of health care. Currently, DOR’s 720-plus staff, including 73 research and staff scientists, are working on nearly 630 epidemiological and health services research projects. For more information, visit divisionofresearch.kp.org or follow us @KPDOR.
Comments (0)