Physics-based Deep Spatio-temporal Metamodeling for Cardiac Electrical Conduction Simulation

Modeling and simulation have been widely used in both cardiac research and clinical study to investigate cardiac disease mechanism and develop new treatment design. Electrical conduction among cardiac tissue is commonly modeled with a partial differential equation, i.e., reaction-diffusion equation, where the reaction term describes cellular excitation and diffusion term describes electrical propagation. Cellular excitation can be modeled by either detailed human cellular models or simplified models such as the FitzHugh-Nagumo model; electrical propagation can be simulated using either biodomain or mono-domain tissue model. However, existing cardiac models have a great level of complexity, and the simulation is often time-consuming. This paper develops a new spatiotemporal model as a surrogate model of the time-consuming cardiac model. Specifically, we propose to investigate the auto-regressive convolutional neural network (AR-CNN) and convolutional long short-term memory (Conv-LSTM) to model the spatial and temporal structure for the metamodeling. Model predictions are compared to the one-dimensional simulation data to validate the prediction accuracy. The metamodel can accurately capture the properties of the individual cardiac cell, as well as the electrical wave morphology in cardiac fiber at different simulation scenarios, which demonstrates its superior performance in modeling and the long-term prediction.