Machine Learning Based Cell Model for fast approximation of cellular action potential to enable clinical translation

Pau Romero, Miguel Lozano, Giada Romitti, Dolors Serra, Ignacio Garcia-Fernandez, Alejandro Liberos, Miguel Rodrigo, Rafael Sebastian
CoMMLab, University of Valencia


Simulations of cardiac arrhythmias have shown great potential to plan and optimize therapies. However, biophysical models are complex and involve a high computational cost that poses a problem for their clinical translations. Alternative methods such as Eikonal based simulations can help to reduce the cost at the expense of not considering an action potential model. In this work, we present a methodology to predict the action potential curves during Eikonal simulations. We first train a model with data obtained from biophysical simulations, and following we test their ability to obtain realistic action potentials, given a cell state and the diastolic intervals. A simulation study shown that this method is able to reproduce action potentials in a tissue slab during rotor activity and different stimulation protocols, avoiding to solve ionic models, and reducing dramatically the computational cost.