Fast parallel image reconstruction with cone-beam ART algorithm based on SIMD technology

Shunli Zhang,Dinghua Zhang,Mingjun Li,Jun Ji
2010-01-01
Abstract:Algebraic reconstruction technique(ART) is an iterative image reconstruction method, which is suitable for nondestructive detection of large industrial components. The shortcomings of ART are large amount of computation and extremely time cost. In order to improve the reconstruction speed of cone-beam ART, this paper presents a fast parallel image reconstruction method. Firstly, we propose a parallel method to calculate the weight coefficients and voxel indexes according to the symmetry of 3D ray in the scanning mode of cone-beam CT, which can get the weight coefficients and voxel indexes of two rays at the same time. Then, we use SIMD technology of Intel CPU, which can load a number of packed data at one time and realize parallel implementation of projection, calculation of image correction and back projection using SSE instructions. Experimental results show that our method is very effective and a reconstruction speedup about 1.5 times can be obtained without any loss of image reconstruction precision.
What problem does this paper attempt to address?