Background Sudden cardiac death (SCD) remains a leading cause of death in the United States. Although implantable cardioverter defibrillators (ICDs) reduce mortality, appropriate therapies are delivered in only 3-5% of sudden cardiac arrests (SCAs) annually, highlighting the need for improved risk identification for SCD and sustained ventricular tachycardia (VT). Cardiac magnetic resonance (CMR) with late gadolinium enhancement (LGE) is prognostic for ischemic cardiomyopathy (ICM) and SCD; however, the recent PROFID trial found that gross scar parameters lacked predictive value. Radiomics enables granular scar characterization from LGE images. Additionally, a deep-learning artificial intelligence electrocardiogram (AI-ECG) risk estimator (AIRE) trained on millions of ECGs predicts cardiac events. In smaller cohorts, functional principal component decomposition (FPCD) produces more interpretable scalar features from ECG waveforms. This study aimed to develop and validate machine learning models to improve VT/SCD risk stratification and guide ICD selection. Methods Left ventricular LGE images from 455 patients with ICM and LVEF 4-49% were analyzed using Pyradiomics to extract ~2,000 scar and myocardial features. Principal component analysis (PCA) was used for dimensionality reduction, and LASSO logistic regression identified PCA components most associated with the primary outcome of sustained VT or SCD. 12-lead ECGs were processed using FPCD to extract scalar parameters characterizing the voltage-time curves, and predictive features were selected via LASSO. Identified predictors were combined with clinical data (LV ejection fraction LVEF and LGE) in a model and evaluated using area under the receiver operating characteristic (AUROC) curves with bootstrapping. Results Among 455 patients with ICM, 59 experienced either SCD or sustained VT; 115 received ICD implants. The combined model demonstrated high predictive accuracy with an AUROC of 0.83 (95% CI: 0.78-0.88). The radiomics-based principal component most associated with VT/SCD was “Contrast,” and higher values were linked to highly heterogeneous LGE signal intensity and increased arrhythmic risk. In parallel, the FPCD analysis of ECGs demonstrated that parameters based on leads aVF and V3 had the best association with VT/SCD. A positive R wave in V3 was associated with the primary outcome. Notably, in the subgroup with LVEF 36-49%, using a model of LVEF, AI-ECG, and CMR radiomics to predict the primary outcome had an AUROC of 0.94 (95% CI: 0.90-0.97). Conclusion Integrating radiomics from LGE images and 12-lead ECG features with LVEF in this model enhances risk stratification in VT/SCD and may fill a critical gap for patients with ICM and LVEF 36-49% who lack established primary prevention ICD indications.