Artificial Intelligence (AI) Detects Heart Transplant Rejection

Combined Model Identifies Stable Transplant Patients Accurately

A computer model combining heart electrical recordings with blood tests identified 94 percent of patients who were not rejecting a transplanted heart. Researchers at New York University (NYU) Langone Health led the work. Their findings appeared in JHLT Open after testing on a validation group of 38 adult transplant recipients.

Standard care requires tissue biopsies. That procedure requires clinicians to insert instruments into a blood vessel to cut away small pieces of heart muscle. Specialists then examine the physical sample under a microscope for inflammatory damage.

Blood Tests Alone Generate Frequent False Alarms

Post-transplant blood panels frequently trigger unnecessary surveillance surgeries by overestimating rejection risk in stable patients. Standard tests track gene expression linked to cellular rejection alongside fragments of donor deoxyribonucleic acid (DNA). These markers often trigger false alarms.

In the test group, the blood-test model mistakenly flagged 19 healthy patients for invasive biopsies. Electrocardiograms (EKGs) record cardiac electrical activity through skin sensors to provide complementary data. Surface tracings clarified ambiguous results.

Training Data Paired Thousands of Tracings with Biopsy Records

Investigators trained the artificial intelligence (AI) program on 5,300 EKG recordings collected from 2,357 adult patients between 2018 and 2024. Each tracing matched a recent biopsy. The software separated cases into mild rejection versus moderate or severe rejection episodes.

Doctors treat moderate and severe cases. Identifying tissue inflammation early allows medical teams to adjust immunosuppressive medications before permanent organ damage develops. Avoiding unneeded surgical procedures prevents mechanical stress on healing cardiovascular tissue.

Broader Multi-Center Testing Remains the Next Step

The diagnostic tool requires validation across multiple independent hospitals before clinicians can consider everyday use. The test cohort included 38 patients. Larger trials are needed to prove whether the model maintains its accuracy across diverse patient populations.

Patients should consult their care team. Transplant recipients should discuss all surveillance plans and medication adjustments directly with their treating physicians.