Multi-GPU based Cluster System for CT Iterative Reconstruction Algorithm

Iterative reconstruction (IR) techniques are able to provide better reconstruction results compared with analytic reconstruction methods in cone-beam CT (CBCT) with incomplete and noisy projection data. However, image reconstruction algorithm remains to be significantly time-consuming especially faced with the large data set which decreases the efficiency of imaging processing. Although graphic processing unit (GPU) has offered an attractive alternative platform to improve the computation efficiency, but the single graphic card is still hard to meet the practical requirement because of the limitation on video memory. This paper presents a distributed parallel method for iterative algorithm using multi-GPU based cluster system, which is designed by algorithm characteristics and hardware structures. The experiments show that the multi-GPU based cluster system is able to achieve the same precision with single node, meanwhile it will gain higher speedup with the increasing number of nodes or GPUs. The computing time of whole reconstruction reduces to almost half of the origin by doubling the computing devices.