基于并行欧式距离变换的三维障碍距离场计算

为了快速计算空间有障碍物的三维欧式距离场,提出一种基于图形处理单元（GPU）的三维空间有障碍物的并行欧式距离场计算算法.该算法包含2步：1) 采用扩展的三维欧氏距离变换算法计算待求物体形状顶点的三维Voronoi图,并用GPU并行计算加速；2) 通过多次迭代快速优化计算有障碍物的欧式距离场.算法在放疗靶区的规划中可用于计算不同级别靶区（GTV）在有障碍物情况下到临床放疗靶区（CTV）的外扩.结果表明,该算法比现有方法一般采用暴力求解的方法性能上有较大提升,同时保持稳定的精度.

Computing 3D distance fields with obstacles based on parallel Euclidean Distance Transform

To compute a 3D distance transform with obstacles swiftly, this paper proposed a novel GPU-based parallel Euclidean Distance Transform method that consists of two stages: 1) a three-dimensional Voronoi diagram was quickly computed by employing an improved parallelized Euclidean Distance Transform algorithm to the underlying object; 2) the Euclidean distance field with obstacles was computed using a multiple iteration process. The proposed algorithm was applied in radiation therapy planning, i.e., computing clinic treatment volume (CTV) from gross treatment volume (GTV). The result shows that our approach preserves the accuracy of the distance field and achieves great performance speedup comparing to existing approaches, which used brute-force solutions.