NYU Researchers Find AI May Help Avoid Unnecessary Heart-Transplant Biopsies

An AI model that combines electrical heart recordings with blood tests may help spare heart-transplant recipients unnecessary biopsies, NYU Langone researchers reported October 6. In a test group of 38 recipients, it correctly identified 94% of patients who were not experiencing rejection. The researchers say the approach warrants testing in more patients at several transplant centers.

The false alarms behind the experiment

A biopsy remains the gold standard for diagnosing heart-transplant rejection. Clinicians remove a small piece of heart muscle and inspect it under a microscope for inflammation and other signs that the immune system is attacking the donated organ.

Two blood tests already help assess rejection risk. They measure gene activity linked to cellular rejection and fragments of donor DNA in the recipient’s bloodstream. Those biomarkers can spot rejection, but frequently produce false-positive results that lead to unnecessary biopsies, the researchers said.

Adding the heart’s electrical signal

The team explored whether electrocardiograms, or EKGs, could sharpen that assessment. An EKG records the heart’s electrical activity through sensors on the skin. Researchers trained AI models to recognize patterns in 5,300 recordings from 2,357 adult heart-transplant recipients.

One model used EKG data alone; another combined the recordings with the two blood-test results. Biopsy records from the same patients provided the reference for checking predictions. For the combined model, researchers used patients treated between 2018 and 2024, matching each EKG to a biopsy performed within the preceding month.

The study grouped biopsy findings into no or mild rejection versus moderate or severe rejection. That distinction matters to treatment: changes such as adjusting immune-suppressing medication are typically made only for the more serious cases. The model was therefore evaluated against these grouped findings, rather than treating every degree of rejection as equivalent.

94%Patients without rejection correctly identified

NYU reported this result for the combined model in an additional test group of 38 heart-transplant recipients.

Potentially fewer biopsies, with wider testing ahead

In the test group, a blood-test-based model incorrectly flagged 19 patients as potentially needing a biopsy. Researchers said the combined model would have spared those patients the procedure. That is a projected benefit from the model comparison, not a report of biopsies actually avoided during patient care.

The 94% figure describes correct identification of patients without rejection; it is not an overall accuracy score or a measure of how many rejection cases were caught. The study was published online September 25, according to Medical Xpress’s NYU-provided account. NYU Langone funded the research.

The next planned step is testing the model in more patients across several transplant centers. That would extend the evaluation beyond the small test group behind the reported result.