摘要
Nvidia从GeForce8系列开始,在显卡上推出统一计算设备框架技术,使GPU的通用计算(GPGPU)从图形硬件流水线和高级绘制语言中解放出来,开发人员无须掌握图形学编程方法即可在单任务多数据模式(SIMD)下完成高性能并行计算。在医学图像分析中,图像配准通常是一个耗时的过程,不利于临床应用,为了加速医学图像的2D-3D配准过程,研究了CUDA的设计思想和编程方式,提出了一种基于CUDA并行编程模型的加速配准新技术,在构建的虚拟X线摄像系统下,采用并行计算的方式快速生成高质量DRR图像,以对应像素的灰度值残差作为相似性测度,使用Powell优化方法寻找最优变换。实验结果表明,该技术既很好地保持了配准精度,同时又大大提高了配准速度,加速比达到了十几甚至几十倍。
The new language of Compute Unified Device Architecture（CUDA） technology by Nvidia has freed Computation on the Graphic Processing Uni（tGPGPU） technology from the graphics fixed pipeline and high-level shader language,allowing the design and implementation of Single Instruction,Multiple Data（SIMD） parallel algorithms on a much more simple way than previous method based on texture rendering.In medical image analysis,image registration generally takes relatively long time,which is not feasible for clinical applications.In order to accelerate the 2D-3D medical image registration process,the core concept of CUDA is studied firstly,and then a new technology which based on CUDA parallel architecture is proposed.Whith this technology,under the proposed virtual X ray camera system,high quality DRR images can be produced fastly,and at the same time,the corresponding pixel residuals is taken as a measure of similarity,Powell optimization method is used to find the best transformation.Experimental results show that this technology not only maintains the registration accuracy but also greatly increases the speed of registration process with the speed ratio of more than a dozen or even several times.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第11期56-59,共4页
Computer Engineering and Applications
基金
广东省产学研项目No.cgzhzd0717
Production and Research Projects in Guangdong Province under Grant No.cgzhzd0717
关键词
医学图像
2D-3D图像配准
数字影像重建
图形处理器
统一计算设备架构
虚拟X线摄像系统
medical image
2D-3D image registration
Digitally Reconstructed Radiograph（DRR）
Graphic Processing Uni（tGPU）
Compute Unified Device Architecture（CUDA）
virtual X ray camera system
