Study on 3 Dimensional registration for pipe-shaped objects with branches

本研究的目的是建立一个利用螺旋CT数据进行肺癌筛查的系统。构建该筛选系统的基本思路来源于三维处理容易从三维螺旋CT数据中提取肿瘤。在该系统中，肿瘤被提取为多颞叶螺旋CT数据之间肺血管三维形状的变化。研究的具体目标是构建肺血管识别子系统，构建多时段螺旋CT数据间肺血管分支匹配子系统，构建肺血管三维形态配准子系统。肺血管的识别：提出了一种基于连续区域生长（SRG）的肺血管分支识别方法。从心脏内部，SRG通过逐层处理来识别表面元素。在血管中，附着在原始层上的体素被视为相邻层。层的重力点被看作是sk的一部分。分支点是在新层由两个或两个以上的岛屿组成时发现的。这意味着原层与新层之间的同伦发生了变化。引入了一种抑制假分支和分支匹配的重要指标。血管分支匹配：首先对多层螺旋CT数据中的肺血管骨架进行配准。最初建立了主分支模型，假设血管只有二元分支，只使用显著性来匹配分支。然而，一些血管在心脏附近有明显的三分支，导致分支的不确定性。为此，提出了一种用于分支匹配的球面搜索算法。球体被放置在先前匹配的分支点对上，它们的半径同步增加，直到其中一个表面到达下一个分支点。计算骨架与球面相交点的相对位置，选取距离最近的对作为分支的匹配对。利用带分支的小型船舶数据对其性能进行了评价。血管配准：提出了一种分段线性变换方法，利用骨架匹配结果在多层螺旋CT数据之间进行血管体的共配准。该方法将数据区域划分为三金字塔，其顶点为分支点匹配。分割后，对相应的三金字塔对进行三维仿射变换。利用SRG从多层CT数据集中识别和提取的人为扭曲的肺血管来评估其性能。少

The objective of the research was to construct a system for lung cancer screening using helical CT data. The basic idea for constructing such a screening system came from that 3 Dimensional processing easily extracted tumor from 3D helical CT data. In this system, tumors were extracted as changes of 3D shape of lung vessels between multi-temporal helical CT data. Concrete targets of the research were construction of sub-system for recognizing lung vessels, one for matching branch of lung vessels between multi-temporal helical CT data and one for registering 3D shape of lung vessels.1.Recognition of lung vessels : A method named as successive region growing (SRG) was developed for the recognition of pipe-shaped object with branches. From the inside of the heart, SRG recognized a surface element by using layer by layer processing. In the vessels, voxels attached to the original layer were regarded as the neighboring layer. The gravity point of the layer was regarded as the part of the sk … More eleton. The branching point was found when new layer consisted of two or more islands. This meant the homotopy changes between the original layer and new one. An index, significance was also introduced to reject false branches and to branch matching.2.Vessel branch matching : Firstly, skeletons of lung vessels among multi-temporal helical CT data were registered. Originally, principal branch model was developed and branches are matched by using only significance on the assumption that vessels had only binary branching. Some vessels, however, had apparent triple branching near the heart and caused uncertainty in branching. So, spherical surface search algorithm was newly developed for the branch matching. Spheres were put on a previously matched branching point pair and their radii were synchronously increased until one of surfaces reached to the next branching point. Relative position of crossing points among skeletons and sphere surfaces were evaluated and nearest pairs were selected as the matched pair of branches. The performance was evaluated by using small scale vessel data with branches.3.Registration of vessels : A piece-wise linear transformation method was proposed for the co-registration of vessel bodies between multi-temporal helical CT data using the skeleton matching results. In the method, the data region was divided into tri-pyramids whose vertices were branching points matched. After the division, 3D Affine transformation was applied to the corresponding pair of tri-pyramids. The performance was evaluated by using artificially distorted lung vessels which were originally recognized and extracted by SRG from a multi-slice CT data set. Less