AI-enabled Vectorcardiography (AI-VCG) combined with cardiac magnetic resonance for the mechanism of left bundle branch block in patients undergoing cardiac resynchronization therapy.
Abstract
Background
Left bundle branch block (LBBB) is a heterogeneous conduction disorder with major implications for cardiac resynchronization therapy (CRT). Conventional ECG criteria incompletely distinguish intact Purkinje activation (IPA-LBBB) from complete conduction block (CCB-LBBB), limiting precision pacing decisions.
Objective
To develop an AI-based, patient-specific vectorcardiographic (AI-VCG) framework that integrates ECG and cardiac magnetic resonance (CMR) to noninvasively identify LBBB mechanisms.
Methods
A patient-specific AI-VCG model was constructed using individualized ECG-to-VCG transformations derived from cardiac-torso geometry and expressed in cardiac-centric coordinates. Performance was evaluated using in-silico electrophysiology simulations (n = 48 hearts) and a clinical CRT cohort (n = 38). A physiologically interpretable AI-VCG feature-the integral of the left-right cardiac vector (IcX)-was assessed for mechanism classification.
Results
In-silico validation demonstrated strong discrimination of IPA-LBBB versus CCB-LBBB (AUROC 0.96). The clinical IcX analysis separated IPA-LBBB and CCB-LBBB with 100% accuracy in the LBBB subset (n = 25), with a 22.4 μWb gap between the highest IPA and lowest CCB IcX values. Direct comparator analyses showed 100% accuracy for Kors-derived QRS area (but with a narrower gap), 76% accuracy for Strauss criteria, and 68% accuracy for sex-specific QRS duration alone.
Conclusions
A patient-specific AI-VCG framework provides an explainable, noninvasive and anatomically grounded representation of LBBB mechanisms. These preliminary results support further validation against conventional ECG and VCG metrics in larger multicenter cohorts.