Key Highlights :
US researchers have developed an AI system, MAARS, that can predict risk of sudden cardiac death with a level of accuracy of up to 93% for high-risk populations.
It combines cardiac MRI scans with patients' histories and performs better than existing clinical guidelines by almost double.
Key Background :
Sudden cardiac death (SCD) is a significant public health concern, usually occurring in people who show no previous symptoms. Among the leading causes is hypertrophic cardiomyopathy (HCM), a genetic heart disorder that makes the heart muscle thick and less effective. In the past, it has been challenging to identify which patients with HCM are at risk of SCD. Current American and European clinical recommendations accurately identify only about 50% of high-risk patients, exposing a large number of patients to the potential for either undue risk or over-treatment with devices such as ICDs.
To help close this gap, Johns Hopkins researchers launched MAARS—a machine-learning model that employs a multimodal approach to maximizing prediction precision. Different from other models, which draw heavily on standalone data points or generic danger signs, MAARS involves the integration of high-resolution imaging through contrast-enhanced cardiac MRIs with patient-specific clinical data, such as age, history, and test results.
The model works through a combination of three deep learning networks. One is a 3D vision transformer that has learned to recognize and understand fibrotic scarring in the heart muscle—a subtle but important sign of arrhythmic risk that could elude human readers. The second network takes structured data from the medical records and discerns patterns of risk that are associated with clinical outcomes. A third fusion layer then combines both sources of information into a single, understandable prediction.
MAARS has also been evaluated using patient cohorts from Johns Hopkins and the Sanger Heart & Vascular Institute and worked extremely well. Particularly impressive is its 93% accuracy rate in adults 40–60 years old, an age group vulnerable to high SCD rates but in which existing risk models are ill-defined.
Aside from accuracy, another useful innovation is the transparency of the model. It not only estimates risk but also points out the exact MRI regions and clinical factors that generate the outcome. This kind of interpretability encourages clinicians to trust it and allows for more rational decision-making.
In the future, scientists intend to modify MAARS for future risk prediction in comparable scenarios like cardiac sarcoidosis and arrhythmogenic right ventricular cardiomyopathy. Additionally, clinical trials in larger cohorts will serve to validate the model in varying population settings.
MAARS is a significant milestone towards precision medicine-based, personalized cardiac care. With the incorporation of state-of-the-art imaging and AI-based analysis, it is a transparent, trustworthy, and potentially life-saving tool for the prevention of sudden cardiac death.