An improved viscous fluid elastic registration algorithm of medical images based on parallel computation
Wang Yangping1, 2, Wang Bing1, Du Xiaogang1
COMPUTER MODELLING & NEW TECHNOLOGIES 2014 18(12A) 354-359
1School of Electronic & Information Engineering, Lanzhou Jiaotong University, 88 An Nin West Road Lanzhou,Gansu, China
2 School of Mathematics and Statistics, Lanzhou University,222 Tian Shui South Road Lanzhou,Gansu, China
Image elastic registration based on the viscous fluid model is time-consuming to solve partial differential equation (PDE) to obtain the velocity and displacement field of the floating image. The method is also sensitive to the gray-scale transformation, with a moving force generated by the gray difference between the two pre-registered images. This paper proposes an improved algorithm which employs the mutual information as a particle moving force, thereby moving the floating image space to obtain the reference image. Conjugate gradient iteration, which is suitable for parallel computing is used to solve the PDE. The proposed registration processing is designed to perform on compute unified device architecture (CUDA) which coordinates CPU and GPU to accomplish concurrent computation. The solution of the PDE is effectively performed on multi-GPU. Experimental results show that the presented method improves the robustness and efficiency of the registration based on the viscous fluid model.