Accelerating the simulation of brain tumor proliferation with many-core GPUs

摘要

Medical centers, such as hospitals, clinics and diagnostic centers, form a special category of facilities, where the need to perform demanding scientific simulations has to be combined with a reasonable deployment cost in order for such simulations to be applicable at a wide scale. Under these circumstances, the use of supercomputing clusters can not be considered as a universal solution. Nevertheless, we argue that the use of the newly introduced multi-core and many-core microprocessors - either at the local level or through cloud computing infrastructure - can lead to significant speedups if the necessary software development effort is expended. In the current paper, in order to give evidence of the feasibility of such an approach, we present a numerical method for the simulation of brain tumors proliferation and we demonstrate the acceleration of this method in the context of a state of the art many-core GPU. The numerical solution is based on the high-order discontinuous Galerkin (DG) method and the simulation is performed on the unstructured mesh that results from the space discretization of the brain volume. Two implementation schemes using CUDA and one multithreaded implementation using OpenMP are evaluated and they highlight the potential speedup that a diagnostic process can experience in a facility that is equipped with a single node multi-core or many-core microprocessor.