A multimodal optimisation framework for cardiac electrophysiology model personalisation using computed tomography and 12-lead electrocardiogram: Algorithm development and validation

Accurate parameterisation of cardiac patient-specific models is a major challenge. Here we aimed to test the feasibility of an optimised multimodal personalisation framework that combines structural information from computed tomography (CT) and 12-lead electrocardiograms (ECGs) features, for high-fidelity simulations of ventricular tachycardia (VT).

We studied a small cohort of six patients who had CT images and sinus-rhythm 12-lead ECGs available prior to VT ablation. The CT images were used to construct patient-specific biventricular simulation domains and to estimate ECG lead positions. We implemented a graphics processing unit-based computational pipeline and a two-stage optimisation framework consisting of: (i) an early activation onset estimation; and, (ii) the optimisation of conductivity parameter. For validation, the simulation outcomes were compared to clinical VT patterns and cycle lengths.

Following early onset estimations, simulated 12-lead ECG signals achieved a $91.20\%$ mean Q wave, R wave, and S wave (QRS) peak accuracy. Using the optimised parameters, average QRS duration error was reduced to $3.94$ ms, compared to $37.42$ ms for baseline parameters. Notably, in four patients with available electro-anatomical VT maps, simulations reproduced the clinical VT patterns and cycle length closer to recordings than those obtained with baseline parameterisation.

Our pilot study demonstrates the feasibility of a robust electrophysiology modelling framework able to perform multimodal personalisation by integrating sinus-rhythm ECGs and cardiac CT images for VT simulations with tractable computational time, paving the way for its clinical translation and a less invasive scar-related VT risk assessment.