AI predicts people will suffer sudden heart death

Current medical guidelines identify patients at risk of sudden death and the situation of flipping coins with 50% accuracy.

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

The AI ​​model developed by Johns Hopkins University researchers, published in Natural Cardiology Research, represents a significant advance in cardiac risk assessment. Unlike traditional approaches that rely on limited clinical factors, MAAR analyzes the full range of patient data, including previously underutilized cardiac imaging that reveals hidden patterns in cardiac scars.

Decode the hidden mode of the heart

“At the moment, we are not protected because they are not protected, and others who endure defibrillators for the rest of their lives are not benefiting.” “Whether we can predict with high accuracy whether patients are at high risk of sudden death.”

Hypertrophic cardiomyopathy affects one in every 200-500 people worldwide and is the leading cause of sudden heart death in young athletes. The disease can cause thickening of the heart muscles and can cause fatal arrhythmia, but predicting which patients are at the highest risk has put cardiologists in trouble for decades.

Breakthroughs from MAARS’ ability to contrast enhanced cardiac MRI images with unprecedentedly refined analysis. When doctors struggle to interpret the original imaging data, the AI ​​system identifies key scar patterns associated with the risk of arrhythmia.

Multimodal analysis reveals new insights

MAARS uses three professional neural networks to collaborate on different types of medical data:

• A 3D visual transformer that analyzes original cardiac MRI images to detect fibrosis patterns
• Neural networks process electronic health records and clinical measurements
• A converged module that integrates insights from all data sources to generate risk predictions
• Attention mechanism highlights which image regions and clinical factors drive each prediction

This comprehensive approach addresses the fundamental limitations in current practice. “People don’t learn deeply on these images,” Trayanova noted. “We are able to extract these hidden information in images that are not usually considered.”

Reality verification across demographics

The researchers tested actual patients at Johns Hopkins Hospital and the Sanger Heart and Vascular Institute in North Carolina. The results are surprising: While the current clinical guidelines achieve approximately 50% accuracy, the overall accuracy of MAARS reaches 89%, and 93% accuracy for patients aged 40-60, which is the highest risk of sudden cardiac death.

It is also important that AI models show fairness among different population groups, avoiding the biased problems that plague many medical algorithms. Traditional clinical tools show that performance varies greatly between male and female patients and across age groups, while MAARS maintains consistent accuracy.

The system also demonstrates its interpretability, a key factor in clinical acceptance. Rather than act as a “black box,” Maars explains their predictions by emphasizing specific image regions and clinical factors that contribute to risk assessment in each patient.

Exceed current limit

Johns Hopkins cardiologist Jonathan Chrispin highlights the clinical significance: “Our study shows that AI models significantly improve our ability to predict the highest risk compared to current algorithms and therefore have the ability to change clinical care.”

These implications go beyond improved predictions. MAARS can help doctors make smarter decisions about implantable cardiac reversal defibrillators, devices that can save lives but can experience infection, failure and inappropriate shock risks when deployed unnecessarily.

This work is based on the 2022 AI model and is based on the team’s previous successes that predict the timing of cardiac arrest in patients with heart attacks. Researchers now plan to expand MAAR to other heart diseases, including cardiac sarcoidosis and right cardiomyopathy with arrhythmia.

For thousands of patients with hypertrophic cardiomyopathy, MAAR offers something cardiologically elusive: precise prediction, which can mean the difference between life and death and is provided with the transparency needed to earn the trust of doctors.

Related