基于CUDA的声辐射力弹性成像算法研究  

CUDA-based acoustic radiation force imaging algorithm

作　　者：曾博[1,2] 雷友诚[1] 王丛知[2] 邱维宝[2] 冯歌[2] 曾成志[2] 杨戈[2] 郑海荣[2] 

机构地区：[1]湖南大学电气与信息工程学院,长沙410082 [2]中国科学院深圳先进技术研究院,广东深圳518055

出　　处：《计算机工程与应用》2015年第18期249-254,265,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.81027006)

摘　　要：声辐射力弹性成像是一种新的测量组织硬度的超声成像方法。不同于其他超声组织弹性成像方法,声辐射力弹性成像能够定量测量组织的弹性模量数值,并且具有对操作者经验依赖性低的特点。然而,由于成像算法数据处理量大,运算时间长,声辐射力弹性成像还无法进行准实时的二维成像。为了获得实时的二维声辐射力弹性图像,提出并实现了一种适合于在GPU上并行计算的声辐射力弹性成像算法。通过与运行在CPU上的原始声辐射力弹性成像算法进行对比,证明在GPU上实现的算法大幅度地提高了运算速度。在自制弹性仿体上,比较了基于GPU和CPU两种算法所成的二维弹性分布图像的质量,结果证明两者的图像质量没有明显差异。Acoustic Radiation Force Imaging（ARFI）is a novel method of tissue stiffness estimation. Compared with other conventional ultrasound methods, ARFI is quantitative and less operator dependent. However, current ARFI still cannot provide quasi-real-time 2D images since its algorithm is data-intensive and time-consuming. In order to obtain real-time2 D ARFI image, a new ARFI algorithm which is suitable for parallel computation on GPU is put forward and accomplished.According to the results, a high accelerate ratio has been obtained between GPU and CPU. Comparison of the quality of stiffness mapping images obtained by both CPU and GPU is also performed on a self-made phantom, and it is found that there is no distinct difference between the images.

关 键 词：二维声辐射力弹性成像 统一计算架构(CUDA) 图形处理器(GPU) 并行算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]