An ECG-based model to distinguish cardiac sarcoidosis from arrhythmogenic right ventricular cardiomyopathy

Arrhythmogenic right ventricular cardiomyopathy (ARVC) and cardiac sarcoidosis (CS) can show significant overlap in their clinical presentation. Differentiation between both is crucial for optimal patient management. Aim of this study is to develop an ECG-based machine learning model to distinguish between ARCV and CS patients. The study population included 100 patients (50 with ARCV and 50 with CS) whose 10-s 12-lead ECG were recorded. A total of 630 ECG standard measurements were extracted and submitted to two features selection steps assessing their redundancy and relevance. Sixteen features satisfied the selection criteria and were used as input of a gradient boosting classifier. A 70-30% stratified split was performed. Model performances reached an accuracy, sensitivity and specificity of 0.867 on the test set, correctly classifying 13/15 patients of both ARVC and CS group. These preliminary results show the possibility of using an ECGbased machine learning model to distinguish ARVC and CS.