CUDA平台下的超声弹性成像并行处理算法

A Parallel Algorithm of Ultrasound Strainimaging Based on CUDA

超声弹性模式成像是新兴高端超声成像系统中出现的新型成像模式,与传统的黑白超,彩超成像模式不同,它能够为临床诊断提供组织器官的硬度信息。弹性成像模式可以帮助医生定性和定量地检测组织的弹性值变化,特别是对一些肿瘤疾病如乳腺癌等的早期检测有巨大的推动作用,因此,这一新型检测手段具有十分重大的临床应用价值。但是弹性成像系统在处理时涉及大量的复杂运算,使其难于在临床实时系统中得到应用,为此文章研究并提出一种基于CUDA(Compute Unified Device Architecture,统一计算设备架构)平台的超声弹性成像模式并行处理算法。算法包括了信号预处理,运动计算,应变估计和图像后处理与显示等处理步骤的并行实现。由弹性体模得到的数据实验表明,基于CUDA的超声弹性成像处理结果与基于CPU的实现相比,不仅可以得到相同质量的显示图像,而且可以取得较大的加速效果,满足实时系统需求,文章的数据测试显示对于256×512的信号数据能够达到63fps的帧率,速度提高了85倍。

Ultrasound strain imaging is a brand new imaging mode in the high-end ultrasound imaging system. It can detect and calculate the stiffness of tissues and organs while the traditional imaging mode can＇t like B-mode and color mode. This new-style imaging mode will help doctors qualitatively detect the changes in tissue elasticity and obtain much more precise quantitative information of the variance of tissue and organ stiffness. Thus, there are significant meanings in clinical application. However, because of the massive computation involved in the current ultrasound elasticity imaging system, it will be hard and costly to implement with the traditional processing platform. In this paper, a new algorithm of ultrasound elasticity imaging based on CUDA（Compute Unified Device Architecture） parallel processing platform is presented. This method mainly includes the following parallel procedures such as pre-processing, motion calculation, strain estimation, post processing and so on. Test results from the elastic phantom data, not only show the output of graphics processing unit（GPU） is definitely the same as the one of CPU, but also demonstrate the obvious speedup using GPU, that is, it can achieve 63fps for the data size（256 beam lines, 512 sample points） which is 85 times faster than the CPU implementation.