AI Predicts Spots People About to Suffer Sudden Cardiac Death todayheadline

Current medical guidelines for identifying patients at risk of sudden cardiac death perform about as well as flipping a coin—roughly 50% accuracy.

Now, a multimodal artificial intelligence system called MAARS has achieved 89% accuracy in predicting which patients with hypertrophic cardiomyopathy will experience life-threatening arrhythmias, potentially saving thousands of lives while preventing unnecessary medical interventions.

The AI model, developed by Johns Hopkins University researchers and published in Nature Cardiovascular Research, represents a major advance in cardiac risk assessment. Unlike traditional approaches that rely on limited clinical factors, MAARS analyzes the complete spectrum of patient data, including previously underutilized heart imaging that reveals hidden patterns in cardiac scarring.

Decoding the Heart’s Hidden Patterns

“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,” explains senior author Natalia Trayanova, a Johns Hopkins researcher focused on AI 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.”

Hypertrophic cardiomyopathy affects one in every 200-500 people worldwide and ranks as a leading cause of sudden cardiac death in young athletes. The condition causes heart muscle thickening that can trigger fatal arrhythmias, but predicting which patients face the highest risk has stumped cardiologists for decades.

The breakthrough came from MAARS’s ability to analyze contrast-enhanced cardiac MRI images with unprecedented sophistication. While doctors struggle to interpret the raw imaging data, the AI system identifies critical scarring patterns that correlate with arrhythmia risk.

Multimodal Analysis Reveals New Insights

MAARS employs three specialized neural networks working in concert to process different types of medical data:

• A 3D vision transformer that analyzes raw cardiac MRI images to detect fibrosis patterns
• Neural networks processing electronic health records and clinical measurements
• A fusion module that integrates insights from all data sources to generate risk predictions
• Attention mechanisms that highlight which image regions and clinical factors drive each prediction

This comprehensive approach addresses a fundamental limitation in current practice. “People have not used deep learning on those images,” Trayanova noted. “We are able to extract this hidden information in the images that is not usually accounted for.”

Real-World Validation Across Demographics

The researchers tested MAARS against real patients from Johns Hopkins Hospital and North Carolina’s Sanger Heart & Vascular Institute. The results were striking: while current clinical guidelines achieved roughly 50% accuracy, MAARS reached 89% overall accuracy and an impressive 93% accuracy for patients aged 40-60—the population at highest risk for sudden cardiac death.

Equally important, the AI model demonstrated fairness across different demographic groups, avoiding the bias problems that plague many medical algorithms. Traditional clinical tools showed significant performance variations between male and female patients and across age groups, while MAARS maintained consistent accuracy.

The system also proved its interpretability—a crucial factor for clinical acceptance. Rather than operating as a “black box,” MAARS explains its predictions by highlighting specific image regions and clinical factors that contribute to each patient’s risk assessment.

Beyond Current Limitations

Co-author Jonathan Chrispin, a Johns Hopkins cardiologist, emphasizes the clinical significance: “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.”

The implications extend beyond improved predictions. MAARS could help doctors make more informed decisions about implantable cardioverter defibrillators—devices that can save lives but also carry risks of infection, malfunction, and inappropriate shocks when deployed unnecessarily.

This work builds on the team’s previous success with a 2022 AI model that predicted cardiac arrest timing in heart attack patients. The researchers now plan to expand MAARS to other heart conditions, including cardiac sarcoidosis and arrhythmogenic right ventricular cardiomyopathy.

For the thousands of patients living with hypertrophic cardiomyopathy, MAARS offers something that’s been elusive in cardiology: precision prediction that could mean the difference between life and death, delivered with the transparency needed to earn physicians’ trust.

Related