AI Detects Heart Transplant Rejection Without Invasive Biopsies

Researchers at NYU Langone Health have developed an artificial intelligence system capable of detecting heart transplant rejection without relying on invasive biopsy procedures. Published September 25 in JHLT Open, the study demonstrates that an AI model analyzing combined electrocardiogram data and blood biomarkers can significantly improve diagnostic accuracy while reducing unnecessary surgical interventions. Current clinical standards require physicians to extract and examine heart muscle tissue under a microscope to identify immune-mediated organ damage. While established blood biomarkers measuring gene activity and donor DNA fragments offer predictive value, they frequently generate false positives, triggering redundant biopsies. To address this clinical bottleneck, the NYU team trained machine learning algorithms on a dataset comprising 5,300 electrocardiogram recordings from 2,357 adult heart transplant recipients treated between 2018 and 2024. The algorithms were cross-referenced with biopsy outcomes recorded within a thirty-day window to validate predictive accuracy. The investigation evaluated two distinct computational approaches. A baseline model processed electrocardiogram patterns in isolation, while an integrated model fused those recordings with standard rejection-risk blood tests. When validated against a separate cohort of 38 transplant patients, the combined model achieved a 94 percent accuracy rate in confirming the absence of rejection. Crucially, the system correctly identified patients who did not require tissue sampling, thereby sparing approximately nineteen individuals from invasive procedures that would have otherwise been recommended based on blood markers alone. Lior Jankelson, associate professor of cardiology and biomedical engineering at NYU, emphasized that electrocardiograms contain dense physiological signals often overlooked in routine monitoring. He noted that early detection remains critical for transplant longevity, as timely immunosuppressant adjustments can prevent organ failure. The study marks the first instance where blood biomarkers and electrocardiogram data are unified within a single AI framework and benchmarked directly against histological verification. Moving forward, the research team will expand validation efforts across multiple transplant centers to assess model performance in diverse clinical environments. The findings underscore a broader trajectory in medical technology where multimodal data integration and machine learning replace or supplement traditional diagnostic workflows, potentially lowering healthcare costs and improving patient outcomes.