Volumetric Segmentation of Complex Bone Structures from Medical Imaging Data Using Reeb Graphs
Volumetric Segmentation of Complex Bone Structures from Medical Imaging Data Using Reeb Graphs
复制标题
使用 Reeb 图根据医学成像数据对复杂骨结构进行体积分割
DOI：
--
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
M. Hermann
中科院分区：
文献类型：
--
作者：
Vitalis Wiens;M. Hermann
在探索医学成像数据集时，通常需要根据不同的材料或结构对图像进行分割。基于模型的算法在寻找包围待分割结构的封闭边界轮廓方面表现出色。然而，像海绵这样的多孔结构具有复杂的拓扑结构，并不表现出唯一的单一闭合边界轮廓。为了能够分割这种复杂的结构，我们提出了一种新的算法框架，该框架基于表示拓扑信息的Reeb图。图中的每个节点对应于特定图像切片中的体素的连通区域，而边表示相邻切片之间的连通区域。从粗略分割开始，在关键节点处对相应的图进行细化，得到的图的连通分量提供最终分割。我们提出了两种识别关键节点的策略，一种是单独基于动态阈值，另一种是基于单个用户指定的预分割。该方法是在193个啮齿动物头骨的MCT扫描数据集上进行评估的，这些头骨被分割为头骨、左侧和右侧下颌骨。
The exploration of medical imaging datasets often requires a segmentation of the images according to different materials or structures. Model-based algorithms excel in finding closed boundary contours enclosing the structure to be segmented. However, porose structures like Spongiosa have a complex topology and do not exhibit a unique single closed boundary contour. In order to enable segmentation of such complex structures we suggest a new algorithmic framework based on a Reeb graph representing the topological information. Each node in the graph corresponds to a connected region of voxels in a specific image slice while edges indicate connected regions between adjacent slices. Starting with a coarse segmentation, the corresponding graph is refined at critical nodes and the resulting connected components of the graph provide the final segmentation. We present two strategies for identifying critical nodes, one solely based on dynamic thresholding and one based on a single user specified pre-segmentation. The approach is evaluated on a dataset of 193 mCT scans of rodent skulls which are segmented into skull, left and right mandible.