锥束CT重建FDK算法的两级并行计算研究

Two-level parallelism of cone-beam CT image reconstruction using the FDK algorithm

三维锥束CT图像的FDK算法重建由于运算量大,在重建高分辨率的图像时,重建所需时间通常难以满足实际应用的需求,集群并行计算是解决上述问题的常用方法之一。在一个SM P集群系统上,采用M P I和P threads两种模型相结合的方法,通过节点之间的消息传递和节点内部的共享内存,实现了FDK算法的两级并行。

Image reconstruction of 3D cone-beam tomography using the FDK algorithm requires huge computational resource. The reconstruction of a high resolution 3D object would usually take a long time, which is intolerable in real applications. Parallel computing using cluster systems is one of the common approaches to solve this problem. The mixed model with both MPI and Pthreads is used to parallelize the FDK algorithm on a SMP cluster. The computation is parallelized in a two-level manner with both Message-Passing among nodes and Shared-Memory within each node. Our simulation results on the SMP cluster system are reported.