基于分布式多图像处理器的医学影像体绘制算法

目的:计算机医学图像处理具有所需计算量和计算涉及数据量巨大的特点,需将高性能计算技术应用于计算机医学图像处理中,以适应计算涉及数据量大和计算速度快的要求.方法:计算统一设备架构(CUDA)技术可使图像处理单元(GPU)进行通用并行计算,使其能够提供强大的计算能力.由于分布式并行计算是实现高性能计算的主要方式,故提出一种实用的分布式并行计算模式,即基于分布式多GPU的并行计算模式,其中多台计算机将使用消息传递接口(MPI)并行编程环境配置成分布式集群系统.结果:以医学影像体绘制算法为例分析该计算模式,使得其算法计算时间较短,计算效率较高.结论:使用MPI集群和GPU进行算法加速,可很好地解决无法实现医学影像的实时三维重建的问题

Volume rendering algorithm of medical image on distributed GPUS

Objective： Medical image processing with computer needs great computation and process lots of data, so high performance computing will be applied to medical image processing, in order to adapt the great computation requirement. Methods： The CUDA （Compute Unifed Device Architecture ） technology introduced from NVIDIA, made the GPU use the general purpose parallel computing, which provide powerful computational capabilities. The computing power was used in the paper. As distributed parallel computing is the main way to realize high performance computing, this paper presented a practical distributed parallel computing model. It is based on distributed multi-GPU （Graphic Processing Unit） of the parallel computing model,where multiple computers using MPI （Message Passing Interface） parallel programming environment configured as a distributed cluster system. Results： Make an analysis of the computing model, which is based on medical image volume rendering algorithm. It is shown that the algorithm takes less computing time and has higher computational efficiency. Conclusion： The problem of real-time 3D reconstruction in medical image has been solved by using the MIP clusters and the GPU to accelerated the effect of this algorithm.