Introduction: Physiologically mechanistic computational models of cardiac electrophysiology (EP) tailored to individual patients, often referred to as cardiac digital twins (CDTs), show high promise in industrial and clinical applications. As their potential builds on the ability to mechanistically represent individual cardiac function, CDTs must be calibrated to observable data, such as the electrocardiogram (ECG) or electrograms (EGMs). However, calibrating such high-dimensional CDT models with limited observations poses a significant challenge.
Objectives: Develop a biophysically detailed yet computationally efficient model of human organ-scale EP that matches the output of high fidelity bidomain models and achieves near real-time performance.
Methods: We have developed a novel re-entrant eikonal solver for computing depolarization and repolarization times that phenomenologically accounts for the most important salient physiological features including restitution effects, curvature-dependence of conduction velocity and electrotonic diffusion, and supports re-entrant activation sequences. Electrophysiological function of the model is encoded in a JSON-based dictionary. Transmembrane voltages can then be recovered from eikonal based activation and repolarization times, and a template prescribing the shape and duration of action potential. Combined with an efficient lead field approach, a lightweight, flexible and compact tool for computing ECGs and EGMs was built. Sampling ranges can be encoded in the JSON dictionary for each parameter, thus facilitating large scale parameter sweeps to support ECG calibration.
Results and Conclusion: We validated the biophysical fidelity of our model by comparing activation and repolarization times, transmembrane voltages and ECGs, under a variety of activation sequences against a high resolution reaction-diffusion bidomain model. We show a very close match between both models for all signals that are indiscernible within the margins of uncertainty in these models. Our novel model implemented in the software Gizmo yields full physics bidomain EP with close to real-time performance, thus opening a wide range of applications that were intractable before.