ARRHYTHMIC risk prediction in ischaemic heart disease may be significantly improved by integrating cardiac magnetic resonance imaging (MRI) markers with machine learning, according to new research.
Sudden arrhythmic death remains a major cause of mortality in patients with ischaemic heart disease, yet current risk assessment relies heavily on left ventricular ejection fraction (LVEF).
Researchers evaluated whether cardiac MRI derived measures of myocardial scarring could provide additional prognostic information beyond conventional clinical markers.
Machine Learning Outperforms Conventional Models
The study analysed two independent cohorts comprising 823 patients in total.
The first cohort included 399 patients and 54 major arrhythmic events, while the second cohort included 424 patients and 50 events.
Clinical variables and cardiac magnetic resonance derived scar characteristics were assessed using Cox proportional hazards regression, Random Survival Forests, and DeepSurv survival models.
Across all model configurations DeepSurv, a machine-learning framework for survival analysis, demonstrated superior discrimination compared with conventional regression methods.
Random Survival Forests also performed well, particularly in pooled analyses, while Cox proportional hazards models remained comparatively stable and interpretable. However, researchers observed that both generally fell behind DeepSurv in predictive accuracy.
Advancing Arrhythmic Risk Prediction
The findings reinforce the importance of scar heterogeneity as a marker of arrhythmogenic abnormalities and suggest that arrhythmic risk prediction should extend beyond global measures of cardiac function.
Imaging markers describing tissue complexity appeared to provide complementary information to traditional variables such as LVEF, ventricular volumes, age, and prior myocardial infarction.
The researchers concluded that combining cardiac MRI derived scar characteristics with machine learning survival modelling could improve identification of patients at higher risk of major arrhythmic events.
Among the methods evaluated, DeepSurv showed the greatest ability to generalise across distinct patient populations, supporting the potential role of advanced machine learning approaches in future arrhythmic risk prediction strategies for patients with ischaemic heart disease.
Reference
Sen A et al. Improving arrhythmic risk prediction using cardiac magnetic resonance within deep learning in ischemic heart disease. npj Cardiovasc Health. 2026;DOI:10.1038/s44325-026-00142-5
Featured image: samunella on Adobe Stock
- Author:
