用通用显卡加速三维锥束T-FDK重建算法  

Accelerated 3-D T-FDK reconstruction algorithm using the commordity graphics hardware

作　　者：戴智晟[1] 陈志强[1] 邢宇翔[1] 张丽[1] 毕文元[1] 

机构地区：[1]清华大学工程物理系,北京100084

出　　处：《清华大学学报（自然科学版）》2006年第9期1589-1592,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(10575059);教育部博士点基金资助项目(20030003074)

摘　　要：利用通用显卡实现算法加速是一种适应CT重建特点的重要硬件加速方法。为了提高三维锥束T-FDK算法的重建速度,在对T-FDK算法简单描述的基础上结合显卡特点提出了快速实现T-FDK算法重建的方法。由于显卡的浮点管道比8bit的纹理光栅化管道慢,但是精度高,将两者结合,在力求达到重建速度和图像质量的平衡的基础上实现了该方法。实验结果表明,与经过初步对称性优化后的T-FDK软件算法相比,快速算法达到了27.6倍的时间加速比,从而在保证图像质量的情况下提高了重建的速度。PC graphics boards graphics process unit （GPU） can significantly accelerate computed tomography （CT） reconstruction algorithms. An accelerated 3-D T-FDK algorithm was developed based on a floating point GPU that is structurally different from the original FDK method. The reconstruction speed and image quality were balanced by using both floating point pipelines, which are slower than hardwired 8-bit texture mapping integer processor but more precise numerically, and the 8-bit integer processor. The GPU accelerated T-FDK method is 27.6 times faster than a software implementation. The results show that this method effectively accelerates the algorithm while guaranteeing image quality.

关 键 词：三维锥束CT T—FDK算法 通用显卡 加速重建 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]