Método de lattice Boltzmann para simulação da eletrofisiologia cardíaca em paralelo usando GPU

Abstract

This work presents the lattice Boltzmann method (LBM) for computational simulations of the cardiac electrical activity using monodomain model. An optimized implementation of the lattice Boltzmann method is presented which uses a collision model with multiple relaxation parameters known as multiple relaxation time (MRT) in order to consider the anisotropy of the cardiac tissue. With focus on fast simulations of cardiac dynamics, due to the high level of parallelism present in the LBM, a GPU parallelization was performed and its performance was studied under regular and irregular three-dimensional domains. The results of our optimized LBM GPU implementation for cardiac simulations shown acceleration factors as high as 500x for the overall simulation and for the LBM a performance of 419 mega lattice updates per second (MLUPS) was achieved. With near real time simulations in a single computer equipped with a modern GPU these results show that the proposed framework is a promising approach for application in a clinical workflow.