Programme Contents

Massive parallelization of cardiac simulations using unconventional processing architectures

Computer simulation and modeling is recognised as a fundamental approach in the field of cardiac electrophysiology. However, a major obstacle often encountered in this field is that considerable computational power is required in order to provide the results within a reasonable time frame.

One approach to solving this problem is the paradigm of general purpose computing on graphical processing units (GPGPU). Graphical processing units (GPUs) are parallel processing hardware architectures that are traditionally employed for 3-D computer graphics. Recently however, they have been adapted and used for general computation due to their superior computation rate in comparison with the traditional sequential CPU.

In this study we have utilized the GPU as a massive parallel hardware architecture to improve the computational performance of a cardiac simulation environment. To harness the computational power of the GPU in this environment, we identified two major computationally intensive tasks in the process of solving cardiac simulation problems: 1) solving a very large system of sparse linear equations raised from finite element discretization of the problem; and 2) solving a group of differential algebraic equations (DAE) that expresses the behaviour of cardiac cells in the discretized elements. Subsequent to this identification, we implemented a GPU-based parallel solver for these computational bottlenecks and integrated these implementations into the selected framework.

To demonstrate the effectiveness of the proposed approach, a realistic case study was selected where the proposed GPU-based system is used to solve cardiac bidomain equations coupled with a complex ionic cell model. The bidomain equations are discretized over a relatively large unstructured grid expressing the structure of a human cardiac ventriculum and the simulations performed using a 390 GPU high performance computing machine.

Numerical results show an order of magnitude improvement in computational performance compared to the CPU-based implementation. In light of these results, as well as the growth rate in computational power, reduction in price and the availability of GPU chips, there is great potential for the GPGPU paradigm to drive cardiac modelling and simulation in the coming years.