Bioengineering Doctoral Student Co-Authors AI Heart Model Paper Published in Nature
A team including a UCLA bioengineering doctoral student has developed the largest-ever artificial intelligence model for interpreting echocardiograms, ultrasound images of the heart. The study was published in Nature.
The model, called EchoPrime, was trained on more than 12 million data points which paired videos of the heart with physicians’ clinical interpretations. The model was then tested on data from five major medical systems. Unlike earlier AI models, which analyzed single views of the heart for a specific symptom or disease, EchoPrime’s large training corpus allows it to integrate multiple filmed perspectives to help identify rarer cardiac diseases.
The model’s speed and accuracy is particularly important because echocardiograms are one of the most common ways physicians view the heart. EchoPrime has the potential to speed up clinic workflows and improve patient diagnostic care. Although the model’s accuracy has already been validated along multiple axes — cardiologists agreed with EchoPrime’s findings as often as they agree with one another — the researchers plan to continue validating the model using newly generated clinical data.
The first author was UCLA bioengineering Ph.D. candidate Milos Vukadinovic. The study’s co-corresponding authors were Dr. David Ouyang, a research scientist and cardiologist at Kaiser Permanente Division of Research and an adjunct assistant professor at Cedars Sinai Medical Center; and Bryan He, a computer science researcher from Stanford. Ouyang is Vukadinovic’s thesis advisor. Collaborators on the study included researchers from Stanford University, UC San Francisco and Kaohsiung Chang Gung Memorial Hospital, Taiwan. Funding for the research came from the National Heart, Lung and Blood Institute of the National Institutes of Health. Cedars-Sinai and Kaiser Permanente have both reported on the study.