CT图象重建的算法优化和代码优化

卷积反投影是CT图象重建的最重要算法作为一种线性处理,其基本算法并不复杂,但由于CT的数据量庞大,使得图象重建的计算十分耗时.国外CT大都采用基于64位CPU的工作站并配以专门为CT建象设计的阵列处理板,这种昂贵的造价数十万元的计算机系统对一般医院用户来说是很大的经济负担.相反我们一直采用通用PC工作站,不仅降低了成本,而且计算机可以得到迅速的升级.这样,为满足医院临床诊断和治疗对建象速度的需要,我们就必须对该算法进行优化,优化的效果也成为CT软件能否适应市场需要,能否最终产品化的关键.本文总结了针对CT图象重建分别在算法层次和代码层次上所做的优化工作.第一,在算法实现的层次上,(1)在反投影算法中,进行内外循环交换,将最内层循环的计算量最大限度地减少,(2)利用反投影算法的对称性,只计算出部分图象数据并对称得到其它图象数据,(3)根据不同的视野范围自动调整反投影的计算量,(4)充分利用FFT的特点,提高实数线性卷积的效率.(5)数据的行列交换,用于提高CPU对数据的存取速度.第二,在代码编程的层次上,充分利用新一代微机CPU--Pentium III提供的新资源、新指令和相应的代码优化工具软件,重新编写算法关键部分的程序代码,以提高程序的实际执行效率.在东大阿尔派全身CT扫描机CT-C2000上的实际测试表明,通过上述算法优化和代码优化,CT图象重建软件的运行速度提高了8倍以上,己达到了医院实际使用的要求.

Algorithm Optimization and Code Optimization of CT Image Reconstruction

Filtered Backprojection is the most important algorithm for CT image reconstruction. As a linear processing, the algorithm is not so complex. The computation of image reconstruction is, however, very time consuming, because of the huge amount of data of CT.Most CT scanners made by foreign companies use workstations with 64-bit CPU together with some array processing board specially made for CT image reconstruction. Such an expensive (costing tens of thousands US dollars) computer system will be very heavy burden for ordinary hospitals. On the contrary, we have been using general-purpose PC workstations, which have not only lowered cost but also been able to be upgraded rapidly. Then, we have to optimize the algorithm to satisfy the need of reconstruction speed from clinical diagnostics and treatment. The effect of optimization is critical for the CT software to be practical.The optimization work in algorithm and in coding is summarized in this paper. First, for algorithm implementation, (1) in the backprojection algorithm, the outer and inner loops are exchanged, making the computation in the innermost loop decreased to minimal; (2) only part of the image data are computed and the rest are can be obtained by the symmetry of backprojection;(3) the computation is automatically adjusted according to different field of view; (4) the efficiency of real data FFT is doubled using complex data FFT; (5) data is transposed to raise the memory accessing of CPU. Second, for coding, using the new resource and new instructions of Pentium III and some optimization tools, we rewrite the codes of critical parts of the algorithm to raise the execution efficiency of the program. It is shown by the practical testing on NEUALPINE's CT-C2000 scanner, that running speed of CT image reconstruction is raised for 8 times, satisfying the requirements from hospitals.