State Key Laboratory of Robotics and System, Harbin Institute of Technology, 150080 Harbin, Heilongjiang, China.
Comput Methods Programs Biomed. 2010 Mar;97(3):199-210. doi: 10.1016/j.cmpb.2009.06.003. Epub 2009 Jul 23.
3D point distribution model (PDM) of subcortical structures can be applied in medical image analysis by providing priori-knowledge. However, accurate shape representation and point correspondence are still challenging for building 3D PDM. This paper presents a novel framework for the automated construction of 3D PDMs from a set of segmented volumetric images. First, a template shape is generated according to the spatial overlap. Then the corresponding landmarks among shapes are automatically identified by a novel hierarchical global-to-local approach, which combines iterative closest point based global registration and active surface model based local deformation to transform the template shape to all other shapes. Finally, a 3D PDM is constructed. Experiment results on four subcortical structures show that the proposed method is able to construct 3D PDMs with a high quality in compactness, generalization and specificity, and more efficient and effective than the state-of-art methods such as MDL and SPHARM.
基于体绘制的三维点分布模型（3D PDM）可以通过提供先验知识应用于医学图像分析。然而，对于构建 3D PDM，准确的形状表示和点对应仍然具有挑战性。本文提出了一种新颖的框架，用于从一组分割的体积图像中自动构建 3D PDM。首先，根据空间重叠生成模板形状。然后，通过一种新颖的分层全局到局部方法自动识别形状之间的对应地标，该方法将迭代最近点（ICP）全局配准和活动表面模型（ASM）局部变形相结合，将模板形状转换为所有其他形状。最后，构建 3D PDM。对四个皮质下结构的实验结果表明，与 MDL 和 SPHARM 等最先进的方法相比，所提出的方法能够以紧凑、通用和特异性的方式构建高质量的 3D PDM，并且更高效、更有效。
