Multi-modal cardiac analysis: Distinguishing cardiac sarcoidosis from arrhythmogenic right ventricular cardiomyopathy through radiomics and ECG features

Differentiation between cardiac sarcoidosis (CS) and arrhythmogenic right ventricular cardiomyopathy (ARVC) presents a significant diagnostic challenge due to overlapping clinical presentations. This study aims to develop a machine learning model integrating radiomic and electrocardiogram (ECG)-based features to distinguish between these conditions. The study included 41 CS and 16 ARVC patients. Three machine learning models were developed using cardiac magnetic resonance radiomic features, ECG-based features, and their integration. Data underwent a 5-fold nested cross-validation for feature selection and classifier optimization, with the optimal model retrained on a random training subset (70%) and then evaluated on the remaining 30% test set. Semi-supervised LASSO and logistic regression were selected by the cross-validation as the best performing combination and the Gini index determined the optimal final feature set. The radiomic and ECG-based models achieved F1 scores of 0.88 [0.83–0.92] and 0.92 [0.88–0.96], respectively, while the integrated model demonstrated superior performance (F1 score: 0.96 [0.92–1]). Explainability analysis identified the radiomic texture parameter glcm_ClusterShade as the most significant feature, followed by two ECG parameters namely P wave peak time (from P onset) in lead V2 and Q wave area in lead V5. This study demonstrates that integrating radiomic and ECG-based data in machine learning models can enhance CS-ARVC differentiation, potentially improving diagnostic accuracy in clinical practice. However, these findings require confirmation in larger cohorts and further validation before their clinical applicability can be established.