IEEE Trans Ultrason Ferroelectr Freq Control. 2017 Apr;64(4):694-705. doi: 10.1109/TUFFC.2017.2661821. Epub 2017 Jan 31.
Our primary objective of this paper was to extend a previously published 2-D coupled subsample tracking algorithm for 3-D speckle tracking in the framework of ultrasound breast strain elastography. In order to overcome heavy computational cost, we investigated the use of a graphic processing unit (GPU) to accelerate the 3-D coupled subsample speckle tracking method. The performance of the proposed GPU implementation was tested using a tissue-mimicking phantom and in vivo breast ultrasound data. The performance of this 3-D subsample tracking algorithm was compared with the conventional 3-D quadratic subsample estimation algorithm. On the basis of these evaluations, we concluded that the GPU implementation of this 3-D subsample estimation algorithm can provide high-quality strain data (i.e., high correlation between the predeformation and the motion-compensated postdeformation radio frequency echo data and high contrast-to-noise ratio strain images), as compared with the conventional 3-D quadratic subsample algorithm. Using the GPU implementation of the 3-D speckle tracking algorithm, volumetric strain data can be achieved relatively fast (approximately 20 s per volume [2.5 cm ×2.5 cm ×2.5 cm]).
本文的主要目的是在超声乳腺应变弹性成像的框架内，将先前发表的二维（2D）耦合子抽样跟踪算法扩展到三维（3D）散斑跟踪。为了克服计算量大的问题，我们研究了使用图形处理单元（GPU）来加速 3D 耦合子抽样斑点跟踪方法。使用组织模拟体模和体内乳腺超声数据测试了所提出的 GPU 实现的性能。比较了所提出的 3D 子抽样跟踪算法与传统的 3D 二次子抽样估计算法的性能。基于这些评估，我们得出结论，与传统的 3D 二次子抽样算法相比，这种 3D 子抽样估计算法的 GPU 实现可以提供高质量的应变数据（即预变形和运动补偿后变形射频回波数据之间具有较高的相关性，以及具有较高的对比度噪声比应变图像）。使用 3D 斑点跟踪算法的 GPU 实现，可以相对较快地获得体积应变数据（大约每个体积 20 秒[2.5cm×2.5cm×2.5cm]）。
