An artificial intelligence (AI) model has shown remarkable accuracy in predicting which patients are most at risk of sudden cardiac arrest. The results have outperformed current clinical methods used by doctors.

It is developed by researchers at Johns Hopkins University with federal funding. This new system combines underutilized heart imaging with a wide array of medical data to focus on uncovering hidden indicators of heart disease. This innovation could lead to life-saving interventions for high-risk patients. It would also reduce unnecessary procedures, such as defibrillator implants, in those at low risk.

“Currently, we have patients dying in the prime of their life because they aren’t protected and others who are putting up with defibrillators for the rest of their lives with no benefit,” said Natalia Trayanova, senior author and a leading AI researcher in cardiology. “We have the ability to predict with very high accuracy whether a patient is at very high risk for sudden cardiac death or not.”

This study was published in Nature Cardiovascular Research.

The main focus of the research is hypertrophic cardiomyopathy (HCM), a common inherited heart disorder affecting between 1 in 200 and 1 in 500 people globally. HCM is a leading cause of sudden cardiac death. It happens especially in young adults and athletes. While many of them with the condition live normal lives, a subset faces dramatically increased risk—yet identifying them has long challenged clinicians.

According to Trayanova, existing U.S. and European clinical guidelines for assessing risk in HCM patients are “not much better than throwing dice,” with only about 50% accuracy in identifying those at true risk.

The new model, called Multimodal AI for Arrhythmia Risk Stratification (MAARS), processes a comprehensive set of patient data—including contrast-enhanced MRI scans of the heart—to pinpoint scarring patterns linked to sudden cardiac death. These fibrosis patterns are critical indicators, but have typically gone unnoticed or unanalyzed by conventional methods.

“People have not used deep learning on those images,” Trayanova explained. “We are able to extract this hidden information in the images that is not usually accounted for.”

In trials involving real-world patients from Johns Hopkins Hospital and the Sanger Heart & Vascular Institute in North Carolina, MAARS achieved 89% overall accuracy, nearly doubling the success rate of current guidelines. Among patients aged 40 to 60, the model reached 93% accuracy, making it particularly effective in identifying those most vulnerable.

Importantly, the AI system not only predicts risk but also explains the reasoning behind each prediction, allowing physicians to create highly individualized treatment plans.

“Our study demonstrates that the AI model significantly enhances our ability to predict those at highest risk compared to our current algorithms and thus has the power to transform clinical care,” said co-author Jonathan Crispin, a cardiologist at Johns Hopkins.

This innovation builds on previous work by Trayanova’s team, which in 2022 introduced another AI model that predicted survival outcomes for patients with infarcts, including estimates of when cardiac arrest might occur.

Looking ahead, the researchers plan to validate MAARS on larger patient populations and adapt the system for other heart conditions, such as cardiac sarcoidosis and arrhythmogenic right ventricular cardiomyopathy.

Original Publication
Authors: Changxin Lai, Minglang Yin, Eugene G. Kholmovski, Dan M. Popescu, Dai-Yin Lu, Erica Scherer, Edem Binka, Stefan L. Zimmerman, Jonathan Chrispin, Allison G. Hays, Dermot M. Phelan, M. Roselle Abraham and Natalia A. Trayanova.
Journal: Nature Cardiovascular Research
DOI: 10.1038/s44161-025-00679-1