CUDA下使用区域增长的超声斑点噪声抑制

Ultrasound Speckle Reduction with Region Growing on CUDA

在数字彩色超声成像系统中斑点噪声是影响超声图像成像品质的重要原因。使用图像局部区域的统计信息可以较好地识别出斑点噪声和组织结构区域,进而使用自适应滤波抑制斑点噪声。但这一处理涉及大量复杂计算,使其难以在临床实时成像系统发挥作用,为此研究并提出了一种基于新兴的高性能并行计算平台Fermi架构GPU(graphics processing unit图形处理单元)的并行斑点噪声抑制处理算法。数据测试结果显示,与基于CPU的实现相比,采用Fermi架构的GPU处理不仅可以得到完全一致和较好的图像去噪效果,而且可以取得较大的加速性能。对512×512的图像数据能够达到65fps的高帧率,速度提高了大约183倍。

In the digital color ultrasound imaging system,speckle noise is the main cause of the decrease of the ultrasound imaging quality.With local statistic information of the ultrasound images,the speckle noise can be distinguished from the tissue structure,which could be used to reduce speckle noise by adaptive filter based on the detection results.However,because of the massive computation involved in this filter technique,it has been the bottleneck for the clinical real-time imaging system.In this paper,a new parallel algorithm of speckle reduction with histogram matching based on Fermil GPU（graphics processing unit） is presented.The test results 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 65fps for the image size（512×512） which is 183 times faster than the CPU implementation.