Personalized computational modelling of the electrical response to cardiac resynchronization strategies

Background In patients with dyssynchronous heart failure, cardiac resynchronization therapy (CRT) is the preferred treatment. As a more physiological alternative to biventricular pacing (BiVP), left bundle branch area pacing (LBBAP) and LBBAP optimized CRT (LOT-CRT) are gaining traction. While LBBAP may correct a pure proximal left bundle branch block, its electrical effects in the presence of intraventricular conduction disease (IVCD) and myocardial scar remain unclear. The purpose of this study is to create a computational modelling framework which can determine the optimal CRT strategy for a specific patient. Methods To demonstrate the modelling workflow, the framework was applied to a single patient-derived dataset. A biventricular geometry was reconstructed from computed tomography, scar regions were segmented from delayed enhancement cardiac MRI and modelled as a mixture of non-conductive and slow-conducting tissue. Fibre architecture and Purkinje network were generated using a rule-based approach. Electrical activation was simulated using an Eikonal model. Body-surface ECGs were computed using lead-field theory. Model parameters, including Purkinje topology, conduction velocities, the fraction of non-conductive scar, and the interventricular conduction delay, were optimized using a Bayesian approach. The calibrated model was used as baseline and compared with simulations of three pacing strategies: conventional BiVP; non-selective left bundle branch pacing (nsLBBP), achieved by pacing the LBB and surrounding myocardium; and LOT-CRT, achieved by simultaneously pacing the nsLBBP site and the LV free wall. Results The RMSE between simulated and measured ECGs was ~0.1 mV, with morphologies and QRS duration matching the measured ECGs, indicating a realistic reconstruction of the intrinsic activation pattern. Notably, optimization resulted in a significantly reduced Purkinje conduction velocity, suggesting underlying diffuse conduction disease. Figure 1 shows the intracardiac activation sequences and corresponding simulated ECGs for baseline (intrinsic), BiVP, nsLBBP, and LOT-CRT. The total left ventricular activation time was 150 ms during intrinsic activation, 129 ms during BiVP, 167 ms during nsLBBP, and 108 ms during LOT-CRT. Simulated QRS durations were 134 ms during intrinsic activation, 106 ms during BiVP, 134 ms during nsLBBP, and 134 ms during LOT-CRT. Conclusion Patient-specific computational modelling in a patient with IVCD, and lateral LV myocardial scar demonstrated that, for this patient, nsLBBP alone resulted in less synchronous electrical activation than BiVP, while LOT-CRT achieved the highest degree of synchrony of the LV. These results demonstrate patient-specific models may help identify the optimal CRT strategy.