Integration of Machine Learning-Derived Radiomic Features from Cardiac Magnetic Resonance Imaging with Traditional Risk Factors to Improve Prediction of Ventricular Arrhythmias in Patients with Non-Ischemic Dilated Cardiomyopathy

Non-ischemic dilated cardiomyopathy (NIDCM) is a leading cause of heart failure and sudden cardiac death, with ventricular arrhythmias representing a major determinant of morbidity and mortality. Current risk stratification strategies rely heavily on left ventricular ejection fraction and clinical variables, which have limited sensitivity and specificity for predicting malignant arrhythmic events. Cardiac magnetic resonance (CMR) imaging provides detailed tissue characterization, particularly through late gadolinium enhancement and parametric mapping techniques, offering valuable insights into myocardial fibrosis and structural remodeling. Recent advances in radiomics and machine learning allow extraction of high-dimensional quantitative imaging features that may capture subtle myocardial heterogeneity and arrhythmogenic substrates not detectable by conventional imaging metrics. Integration of these radiomic features with traditional clinical, electrophysiological, and biomarker-based risk factors has the potential to significantly enhance predictive modeling for ventricular arrhythmias in NIDCM patients. This work aims to evaluate the incremental prognostic value of machine learning–derived CMR radiomic features when combined with established risk predictors. The study will assess model performance using discrimination, calibration, and reclassification metrics, and explore the potential role of artificial intelligence–driven imaging biomarkers in guiding implantable cardioverter-defibrillator decision-making and personalized therapeutic strategies. This integrated precision cardiology approach may improve early identification of high-risk patients and optimize resource allocation and clinical outcomes in NIDCM.