A new fitting pipeline for cardiac electrophysiology successfully reproduced the dynamical behavior and pattern coherence of 8 realistic human models while increasing computational speed.
A novel fitting pipeline for cardiac electrophysiology models creates simplified surrogate models that faithfully reproduce the complex dynamics of realistic human models, enabling faster digital twin simulations.
Abstract Personalized mathematical models of heart electrophysiology (EP), also known as digital twins, are promising diagnostic and prevention tools that pose no risk to patients. These models require the calibration of a large number of parameters whose values are determined through a specific fitting strategy or pipeline. The fitting pipeline depends on the objective we are pursuing, the mathematical structure of the model in question, and the available measured data. The choice of data input for any fit will determine the features of the system that are reproduced. In cardiac EP people often use the cardiac action potential (AP) morphology, and a single restitution curve (a relationship of how long the heart is in contraction compared to how much rest) serves as the only data source. This can result in models reproducing only a limited set of features and can generate parameter overfitting, which could introduce spurious physiological and dynamical behavior in the final model. We present a new fitting pipeline that incorporates restitution relationships for the AP duration, conduction velocity, activation time, and refractory period. These observables are commonly measured in experimental and clinical setups and provide a local and tissue-level description of the wavefront, activation, waveback, and speed of AP signals; therefore, avoiding the need for importing data from other sources or performing slow and potentially risky procedures in patients. The model we fit is a simple phenomenological model with a low number of parameters, which allows us to create a one-to-one map between the observables and most of its parameters. This strategy reduces overfitting and simplifies the physiological interpretation of the model. Our results show that the pipeline is capable of reproducing important characteristics of the restitution relations (Fig. 1) and the dynamical complexity of realistic human models in single-cell and tissue (Fig. 2). We applied our fitting pipeline to 8 realistic mathematical models from the atria and ventricle. These models are meant to emulate the complexity and variability of human EP (3 of these models have been approved by the Food and Drug Administration for proarrhythmic risk assessment 1). Our results show that the fitting pipeline produces surrogate models with the same level of dynamical behavior, pattern coherence, and fractionation as their realistic model counterparts. In Fig. 2 we show the comparison of the time evolution of the voltage propagation of one of our realistic models and its surrogate fit. The image shows that the surrogate model (bottom) provides a faithful reproduction of the realistic one (above). Finally, the simple structure of these surrogate models allows us to increase the computational speed of the digital twin simulations, bringing us one step closer to a practical real-time digital twin solution.Fig. 1For image description, please refer to the figure legend and surrounding text. Fig. 2For image description, please refer to the figure legend and surrounding text.
Velasco-Perez et al. (Fri,) conducted a other in Heart electrophysiology. New fitting pipeline incorporating restitution relationships vs. Realistic mathematical models was evaluated on Reproduction of restitution relations and dynamical complexity. A new fitting pipeline for cardiac electrophysiology successfully reproduced the dynamical behavior and pattern coherence of 8 realistic human models while increasing computational speed.