AI Measures Heart Scans in Seconds, and Its Numbers Predict Survival Slightly Better Than Doctors' Do
A study of more than 27,000 heart patients in Britain and the United States found that an AI program measuring how well the heart pumps did slightly better than experienced doctors at pointing to who was most at risk of dying. The program takes 2 to 3 seconds per patient. The advantage is small, the study looked back at past records, and doctors would still need to check the results.
Researchers in London published a study today in the medical journal The Lancet Digital Health that compares an AI program with experienced doctors at a common job in heart care: measuring from an MRI scan how strongly the heart pumps.
The measurement is called ejection fraction. It is the share of blood that the heart's main pumping chamber pushes out with each beat, and the study's lead author, George Joy, calls it "one of the most important measurements we make in cardiology." To get the number, the chamber has to be outlined on the scan's images. The team's earlier work found that this typically takes an expert about 13 minutes per patient. The AI program takes 2 to 3 seconds.
The team, from City St George's, University of London and University College London, went back through the records of 27,382 people who had a heart MRI for a known or suspected heart problem. Most were referred to a heart center in London, and about 1,300 were seen in Pittsburgh. Over a typical follow-up of about five and a half years, 3,458 of them died. The researchers asked which number, the doctors' or the program's, did a better job of pointing to the patients most at risk.
The program's did, by a small margin. Once the patients' age and sex were taken into account, the difference was very small, and the authors themselves call the advantage marginal. So this does not show a program seeing something doctors can't. It shows that a measurement made in seconds held up at least as well as one made by a specialist in minutes. "AI is not replacing clinicians, but it can provide faster, more consistent measurements at low-cost," said Rhodri Davies of University College London, the study's senior author.
There are important limits. The study looked back at past records, so it cannot say whether patients do better when the program is used in their care; the authors say that needs studies that follow patients forward. The program made errors for about 1 percent of patients, and the authors say doctors should check its results. Davies invented the tool, and several of the authors hold shares in or are paid consultants to a company called MyCardium AI, a connection they disclose in the paper. Next, the team plans to combine the scan measurements with other tests, particularly the common heart-rhythm test known as an ECG, to catch heart muscle problems earlier.
Sources
- City St George's, University of London — AI measurements of heart scans (news release, via EurekAlert)eurekalert.org · primary
- PubMed — The Lancet Digital Health paper: abstract and disclosureseutils.ncbi.nlm.nih.gov · primary
- Medical Xpress — report on the studymedicalxpress.com