抄録
This paper proposes a parallelization method for brain tumor resection simulation using General Purpose Graphics Processing Unit (GPGPU). The simulation is performed by using corotational FEM for fracture handling. However, the calculating cost is too high to execute in real-time. The bottlenecks of the simulation are solving linear system of equation and assembly procedure of global stiffness matrix. In order to achieve real-time simulation, we apply the large scale parallelization using GPGPU to the bottlenecks. The proposed method optimizes memory consumptions and calculations by taking account of symmetry property of the stiffness matrices. This paper shows that the proposed method enables the simulation to run in around one and half times faster than the case of multi-core CPU parallelization.
