Strict left bundle branch block diagnose through explainable artificial intelligence

Abstract

This study explores the use of SHapley Additive exPlanations (SHAP) values, a machine learning technique, to validate and refine electrocardiographic criteria for strict Left Bundle Branch Block (LBBB). The research utilizes a 1D convolutional neural network (CNN) model to analyze a database of heart failure patients, including those with strict LBBB, non-strict LBBB, no LBBB, and a healthy control group. The model’s performance was evaluated using five classification schemes, with an accuracy exceeding 81% in all cases. The study found that lead V3 emerged as one of the most valuable leads in the classification task across all proposed combinations, a surprising result given its lack of prominence in clinical LBBB diagnosis. This finding suggests that the link between V3 and LBBB, unexplored until now, warrants further investigation. The study concludes that the integration of SHAP values with traditional electrocardiographic analysis can enhance clinical decision-making and optimize patient care in the context of LBBB.