Project Details

Artificial intelligence detection of pediatric heart transplant rejection from echocardiograms

Henry Foote, MD

Summary

BACKGROUND: Pediatric heart transplantation rejection causes morbidity and mortality. Current rejection surveillance is primarily through periodic heart catheterization and endomyocardial biopsy, but this strategy may result in delayed rejection diagnosis and requires an invasive procedure. An improved method for predicting rejection is urgently needed to both provide earlier recognition of rejection and reduce unnecessary invasive procedures. GAP: Non-invasive echocardiograms are routinely performed to monitor children with heart transplantation, but human interpretation of echocardiograms has limited accuracy for rejection. Artificial intelligence (AI) models can accurately interpret heart function from echocardiograms; however, no models have been developed for children with heart transplantation. HYPOTHESIS: A deep learning AI model will accurately predict heart transplant rejection from echocardiograms in children METHODS: We will use a cohort of children with a heart transplant and an echocardiogram and biopsy on the same day at a single institution to train an AI model to predict heart transplant rejection. We will test the model’s ability to predict probability of rejection from a single echocardiogram as well as from serial echocardiograms over time. We will confirm out findings on a prospectively enrolled cohort. RESULTS: Pending. IMPACT: A strong AI model would transform care for children with heart transplant by allowing for more frequent, consistent, and non-invasive screening for rejection. This would have the potential to reduce invasive procedures for low-risk children while also facilitating earlier rejection detection and treatment, resulting in improved outcomes.