New AI model detects heart transplant rejection without biopsies

By analyzing heart rhythm recordings and blood tests, artificial intelligence may accurately flag when a transplant patient's body starts attacking a donated heart, a new study suggests.

The current gold standard for diagnosing heart transplant rejection is a biopsy, the surgical removal of a small piece of heart muscle, which experts inspect under the microscope for inflammation and other changes that may indicate that the body's immune system is rejecting an organ.

Led by NYU Langone Health researchers, the study explored an alternative method: whether combining electrocardiograms (EKGs)-which record the heart's electrical activity using sensors placed on the skin-with blood tests could help identify rejection without the need for an invasive biopsy in many cases.

The team trained AI models to recognize patterns in 5,300 EKG readings taken in 2,357 adult heart transplant recipients. One AI model was trained on EKG data alone, while another AI tool looked at EKG readings and the results of two blood tests commonly used to predict rejection risk. Biopsy records paired with EKG readings from the same patients served as a measure for the tools' prediction accuracy. 

In a test group of an additional 38 male and female heart transplant recipients, the researchers found that the combined model performed better, correctly identifying 94 percent of patients who were not experiencing rejection. By contrast, the model based on blood tests incorrectly flagged 19 patients as potentially needing a biopsy. The combined model, the researchers said, would have spared the patients from the procedure.