Load-Prediction Parallelization for Computer Simulation of Electrocardiogram Based on GPU

This paper introduces a parallel algorithm using GPU for computer simulation of Electrocardiogram (ECG) based on a 3-dimensional (3D) whole-heart model. The computer heart model includes approximately 50,000 discrete elements (cell models) inside a torso model represented by 344 nodal points with 684 triangular meshes. Since computational burden for computer simulation of ECGs is considerably heavy, we employ GPU to accelerate the speed of calculation. However, GPU is based on SIMD structure which is unsuited for branch structure, so that the computing capabilities of GPU are limited by the branch of program. In order to solve this problem, we present a GPU-based algorithm which concentrates on eliminating branches in computation and optimizing the calculation of electric potentials through the way of load-prediction. The new parallel algorithm accelerates the speed of calculation of ECGs to 6.18 times compared with the former algorithm. This study demonstrates an effective algorithm based on GPU for parallel computing in biomedical simulation study.