A GPU-Based Parallel Computing Framework for Accelerating the Reconstruction of q-Ball Imaging

摘要

High-angular resolution diffusion imaging (HARDI) relative to the diffusion tensor imaging (DTI) can resolve the complex fiber crossing of each voxel in the brain, however, the image reconstruction time is longer than conventional technology, therefore, to improve the performance is a very important work to do. We used graphic processing unit (GPU) and CUDA to compute the spherical harmonic function on the reconstruction of QBI. It's very useful on massive performance data (Bid Data). With the large number of matrix elements in GPU for parallel computing. By evaluating with different video cards, the improved performance showed 179 to 574 speed up through shared memory access for computation. Finally, the results will be used in an extension of the three-dimensional spatial visualization and probabilistic tractography.