Probabilistic calibration of a closed-loop cardiac electromechanical model with application to cardiac resynchronization therapy

Cardiac conduction disorders such as left bundle branch block induce ventricular dyssynchrony and increase the risk of heart failure. Cardiac resynchronization therapy (CRT) improves cardiac function in approximately 70% of patients; however, identifying the optimal pacing strategy for a given patient remains challenging. We present a computational framework that integrates physics-based modeling with Bayesian personalization to simulate cardiac electromechanical function and response to CRT. Patient anatomies are reconstructed from cardiac magnetic resonance (CMR), and electrical activation is modeled using a 3D Eikonal formulation with probabilistic inference of the Purkinje network from ECG data via Bayesian optimization. This approach enables uncertainty quantification by identifying multiple activation patterns consistent with clinical observations. Electrical activation is coupled to a closed-loop cardiovascular model, calibrated using constrained Bayesian optimization to match CMR-derived volumetric measurements. The framework enables simulation of multiple pacing strategies and assessment of their acute hemodynamic effects while propagating electrophysiological uncertainty to mechanical outputs. As a proof of concept, the framework is evaluated in four heart failure patients spanning different clinical phenotypes, including differences in sex, cardiac dimensions and myocardial scar. Relying exclusively on non-invasive data and maintaining low computational cost, the proposed framework provides a scalable approach toward uncertainty-aware cardiac digital twins as a potential decision-support tool for evaluating personalized CRT strategies.