Li Min, Xiang Zhikang, Xiao Liang, Castillo Edward, Castillo Richard, Guerrero Thomas
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China.
Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.
Proc IEEE Int Conf Prog Inform Comput. 2015 Dec;2015:292-295. doi: 10.1109/PIC.2015.7489856. Epub 2016 Jun 13.
Deformable registration (DR) is a key technology in the medical field. However, many of the existing DR methods are time-consuming and the registration accuracy needs to be improved, which prevents their clinical applications. In this study, we propose a parallel block matching algorithm for lung CT image registration, in which the sum of squared difference metric is modified as the cost function and the moving least squares approach is used to generate the full displacement field. The algorithm is implemented on Graphic Processing Unit (GPU) with the Compute Unified Device Architecture (CUDA). Results show that the proposed parallel block matching method achieves a fast runtime while maintaining an average registration error (standard deviation) of 1.08 (0.69) mm.
可变形配准（DR）是医学领域的一项关键技术。然而，现有的许多DR方法耗时较长且配准精度有待提高，这阻碍了它们在临床上的应用。在本研究中，我们提出了一种用于肺部CT图像配准的并行块匹配算法，其中将平方差度量之和修改为代价函数，并使用移动最小二乘法来生成全位移场。该算法在具有统一计算设备架构（CUDA）的图形处理器（GPU）上实现。结果表明，所提出的并行块匹配方法在运行时速度很快，同时保持平均配准误差（标准差）为1.08（0.69）毫米。
