Article Impact Level: HIGH Data Quality: STRONG Summary of Nature Cardiovascular Research, 4(7), 891–903. https://doi.org/10.1038/s44161-025-00679-1 Dr. Changxin Lai et al.
Points
- Researchers developed a deep learning model called MAARS to more accurately predict sudden cardiac death in patients with hypertrophic cardiomyopathy, a leading cause of mortality in the young.
- The innovative AI system analyzes multimodal data, including electronic health records, and uniquely incorporates raw contrast-enhanced cardiac MRI images to identify previously hidden risk factors, such as scarring patterns.
- MAARS achieved an area under the curve of 0.89 in an internal cohort and 0.81 externally, significantly outperforming current clinical guidelines by a margin of up to 0.35.
- Unlike traditional methods, the model demonstrates fairness across all demographic subgroups and provides interpretable predictions to help clinicians tailor personalized patient treatment plans for high-risk individuals.
- This breakthrough has the potential to save lives by improving patient selection for preventive therapies, such as defibrillators, and will be expanded to other challenging heart conditions in the future.
Summary
Researchers have developed a deep learning approach, known as MAARS (Multimodal Artificial Intelligence for Ventricular Arrhythmia Risk Stratification), to enhance the prognostication of sudden cardiac death in patients with hypertrophic cardiomyopathy (HCM). This common inherited disease, affecting one in every 200 to 500 individuals, presents a significant challenge in risk stratification, as current clinical guidelines demonstrate low and inconsistent accuracy. MAARS was designed to overcome these limitations by integrating a full spectrum of medical data to provide a more precise, individualized risk assessment for lethal arrhythmia events, which are a leading cause of mortality in this patient population.
The model utilizes transformer-based neural networks to learn from multimodal data sources, including electronic health records, echocardiogram and radiology reports, and, uniquely, contrast-enhanced cardiac magnetic resonance (ceCMR) images. By analyzing the raw ceCMR images, MAARS can identify subtle patterns of fibrosis and scarring that are critical for risk assessment but are often underutilized in clinical practice. The model achieved an area under the curve (AUC) of 0.89 (95% confidence interval [CI] 0.79–0.94) in an internal validation cohort and 0.81 (95% CI 0.69–0.93) in an external cohort. This represents a significant improvement of 0.27–0.35 (internal) and 0.22–0.30 (external) over existing clinical guidelines.
Beyond its superior predictive power, MAARS demonstrates fairness across demographic subgroups, a notable advantage over current standards. The model is also designed for interpretability, enabling clinicians to understand the factors that drive a specific patient’s high-risk prediction, thereby promoting AI transparency and facilitating personalized treatment planning. This approach has the potential to enhance patient selection for therapies like implantable defibrillators and may be expanded to other conditions, including cardiac sarcoidosis and arrhythmogenic right ventricular cardiomyopathy.
Link to the article: https://www.nature.com/articles/s44161-025-00679-1
References Lai, C., Yin, M., Kholmovski, E. G., Popescu, D. M., Lu, D.-Y., Scherer, E., Binka, E., Zimmerman, S. L., Chrispin, J., Hays, A. G., Phelan, D. M., Abraham, M. R., & Trayanova, N. A. (2025). Multimodal AI to forecast arrhythmic death in hypertrophic cardiomyopathy. Nature Cardiovascular Research, 4(7), 891–903. https://doi.org/10.1038/s44161-025-00679-1
