Wang H, Jaszczak R J, Coleman R E
Department of Radiology, Duke University Medical Center, Durham, NC 27710.
Phys Med Biol. 1993 Sep;38(9):1235-62. doi: 10.1088/0031-9155/38/9/005.
A composite object model is proposed for Monte Carlo simulation of radiological imaging systems. The composite model contains four components: a set of regular and 'voxelized' primitives, a 'modular' inclusion tree, a set of designated constructive solid geometry (CSG) trees, and a mapping from the set of CSG trees to the inclusion tree. The voxelized primitive is a primitive containing a stack of voxels whose intersections with a photon path are calculated based on Siddon's method. The inclusion tree is employed to describe the inclusion relationships of homogeneous subregions of material characteristics in larger regions in an object. The model is designed so that the 'divide-and-conquer' principle for modular software design can be used to construct an inclusion tree for a complex object. The designated CSG trees are used to model source distributions. The mapping from the inclusion tree to the CSG trees provides the fundamental information for the initial location of a photon history in the tree. Computational issues are addressed for the new model. For the modular inclusion tree, a computationally efficient algorithm is proposed in conjunction with the determination of the photon-ray intersections with primitive and voxel boundaries as well as with the identification of the material characteristics for each ray segment between two adjacent intersections. For source distributions, the designated CSG trees are defined based on the set intersections-difference-unions (IDU) operation sequence. The IDU operations present computation advantages over existing source modelling based on a single inclusion tree. For voxelized primitives, a new sampling technique called the fractional photon emission technique is introduced to reduce sampling computations for existing techniques based on single photon emission. The composite model is an extension of existing simple-geometry, solid-geometry, and voxel-based models and hence provides greater flexibility for model optimization. Based on the composite model, two objects were simulated: an anthropomorphic thorax with lesion inserts for transmission studies and a 3D Hoffman brain phantom with lesion inserts for emission studies. The simulations demonstrate the flexibility of the composite approach in modelling objects for clinically significant applications.
提出了一种用于放射成像系统蒙特卡罗模拟的复合物体模型。该复合模型包含四个组件：一组规则的和“体素化”的基元、一个“模块化”包含树、一组指定的构造实体几何（CSG）树，以及从CSG树集到包含树的映射。体素化基元是一个包含一堆体素的基元，其与光子路径的交点基于西登方法进行计算。包含树用于描述物体中较大区域内材料特性均匀子区域的包含关系。该模型的设计使得模块化软件设计的“分治”原则可用于构建复杂物体的包含树。指定的CSG树用于对源分布进行建模。从包含树到CSG树的映射为光子历史在树中的初始位置提供了基本信息。针对新模型解决了计算问题。对于模块化包含树，结合确定光子射线与基元及体素边界的交点以及识别两个相邻交点之间每个射线段的材料特性，提出了一种计算效率高的算法。对于源分布，基于集交 - 差 - 并（IDU）操作序列定义指定的CSG树。与基于单个包含树的现有源建模相比，IDU操作具有计算优势。对于体素化基元，引入了一种称为分数光子发射技术的新采样技术，以减少基于单光子发射的现有技术的采样计算。该复合模型是现有简单几何、实体几何和基于体素的模型的扩展，因此为模型优化提供了更大的灵活性。基于该复合模型，模拟了两个物体：一个用于透射研究的带有病变插入物的人体胸部模型，以及一个用于发射研究的带有病变插入物的三维霍夫曼脑模型。模拟结果证明了复合方法在为临床重要应用对物体进行建模时的灵活性。
