Researchers from University of California, Berkeley, Beth Israel Deaconess Medical Center, and the University of California, San Francisco published a study that revealed how AI can analyze heart scans more accurately and quickly than humans.
According to the team, they trained a computer to evaluate the most common echo views with over 180,000 echocardiogram images. Then, they tested both human technicians and the computer on the new set of samples. They said that humans were 70.2 to 83.5 percent accurate in assessing the echo videos, while AI was 91.7 to 97.8 percent accurate. Researchers also claimed that echocardiogram interpretation can be complex because still images, heart recordings, and various video clips are measured from more than a dozen views. With only slight differences in between the views, it is very challenging for humans to provide a standardized and accurate analysis.
The study explains that the use of deep learning has been highly successful in learning the patterns. “Imaging is a critical part of a medical diagnosis,” the study reads. Furthermore, the researchers said that this type of diagnosis can be applied in fields of dermatology, pathology, and in radiology. They said that although AI has not been widely used to echocardiograms because of the complexity, they have gathered randomly selected echo images for training, validation, and testing; and researchers have built a view classification on their own.
UCSF Medical Center’s cardiologist Rima Arnaout said, “This is providing a foundational step for analyzing echocardiograms in a comprehensive way.” The team also admits that they have a long way to go before AI can take the center stage in a clinical setting.