My Research

Surgical simulators are valuable tools for surgical planning and training, and their usability is improving dramatically due to advancement in hardware and simulation techniques. One of the main challenges in this domain is the modeling of deformable objects. Early approaches to modeling deformable objects from Continuum Mechanics, or even Finite Element Methods can not be implemented on a CPU in real time for a high number of polygonal meshes. Faster methods like Mass-Spring systems are not realistic enough and additional constraints are required for better simulation which makes them computationally expensive. Moreover, pre-computed method which are faster and capable of modeling large deformations, are unable to model a wide range of object interactions which are essential for surgery simulation. For specific surgical scenarios hybrids of these methods can be designed, yet still cannot be run in real time on workstation CPUs. Fortunately, NVIDIA's CUDA is an affordable technology enabling users to take advantage of GPU's parallel processing capability. For a typical parallel problem, use of the CUDA can easily result in ten times performance increase. Deformable object modeling is a highly parallelizable problem, therefore great improvements can be expected if impregnated using the Compute-Unified Device Architecture("CUDA").

We have also developed a new method for modeling deformable object which is even faster and more stable than mass spring system.