high performance computing on graphics processing units: hgpu.org

Realism and real-time visual and haptic interactions with anatomical structures are key challenges in simulation software for surgeries. Overcoming these challenges is made difficult by the need to run the software on consumer-grade computing platforms. This paper presents preliminary work towards a framework for fast, realistic and stable simulation of deformable anatomical structures. The approach is to use implicit Euler’s integration method to compute the deformations of the anatomy. This method is reformulated to exploit the inherent parallelism in the resulting simulation equations using Graphics Processing Unit (GPU) -a chip in many mid-range graphics cards. A dynamic surface mesh models the anatomy and a sparse, large linear system solved at each time step of the simulation models the deformations. The inverse of the large sparse matrix is computed as an approximation. The resulting large matrix manipulations are implemented on an NVIDIA GEForce 6600 GT GPU by using textures for storing mesh values and fragment-level programming for computing the new values due to deformations. The resulting simulation is compared with an explicit euler’s integration-based simulation run on the CPU.