《基于水平集主动轮廓模型的医学图像分割方法的研究》由会员分享，可在线阅读，更多相关《基于水平集主动轮廓模型的医学图像分割方法的研究（62页珍藏版）》请在金锄头文库上搜索。
1、合肥工业大学 硕士学位论文 基于水平集主动轮廓模型的医学图像分割方法的研究 姓名：任靖 申请学位级别：硕士 专业：信号与信息处理 指导教师：蒋建国;詹曙 2011-04 基于水平集主动轮廓模型的医学图像分割方法的研究基于水平集主动轮廓模型的医学图像分割方法的研究 摘摘 要要 医学图像分割是对正常组织和病变组织进行定量分析、定性识别的基础， 是计算机辅助诊断的关键步骤，因此要求很强的分割目标性和很高的分割准确 性。然而，在临床应用中，由于医学图像本身的复杂性、多样性和各种不确定 因素的影响，导致准确分割成为医学图像分析处理的难点。 基于水平集主动轮廓模型的图像分割方法，既可以利用低层的图像信息，
2、 又可以结合高层的目标先验知识，因此在处理结构复杂多样的医学图像时，显 示出其独特的优越性和广泛的适用性。本文以骨关节核磁共振图像（MRI）为 研究对象，对基于水平集的几何主动轮廓分割算法提出了几点改进，并在骨关 节正常骨骼和病灶分割实验中验证了改进模型的有效性。具体工作如下： （1）提出一种局部自适应的变分水平集模型。相对于 Li 的无需重新初始化 的水平集模型，主要有两点改进：一是利用图像的局部特性自适应决定曲线的 演化方向，能较好的解决初始化曲线跨越目标边界的分割问题；二是将全局 Chan-Vese 能量项局部化，改进边界停止函数，提高了对灰度分布重叠、分布 不均匀问题处理的鲁棒性，增强
3、了算法的适应性。最后结合医学序列图像特点， 利用 Heaviside 函数对当前图像分割结果进行分段常量化后投射至相邻图像作 为初始化曲线，实现对序列图像的自动分割。完成了对骨关节 MRI 序列图像中 骨肿瘤目标的自动分割。 （2）膝关节 MRI 图像结构复杂，各软组织之间对比度低，分界模糊，给关 节软骨的分割带来了很大的难度，只利用低层图像信息的水平集方法往往难以 得到较满意的分割结果。本文提出一种结合目标先验平均形状信息的水平集分 割模型，利用点分布模型对目标样本轮廓进行先验形状归一化，将平均形状表 示成水平集函数，在局部自适应轮廓模型的基础上，将形状信息融入到水平集 的框架下，实现在迭代
4、过程中对演化曲线的形变进行指导和约束。该算法提高 了对弱边界的识别度，减弱了非目标轮廓的干扰，具有良好的抗噪能力。完成 了对膝关节 MRI 软骨组织的初步分割。 关键词：关键词：医学图像分割；主动轮廓模型；水平集；局部自适应；先验形状 Study on Medical Image Segmentation Based on Level Set Active Contour Model Abstract Medical image segmentation is not only the foundation of quantitative analysis, qualities identif
5、y of the normal and diseased tissues, but also the key step of computer aided diagnoses. Thus it should have an unambiguous segmentation target and requires high segmentation accuracy. However, due to the complexity, diversity, and many other uncertain effects on medical images, accurate segmentatio
6、n is a difficult topic in Processing and Analysis of Medical Images. The level-set active contour model-based image segmentation method, by combining of the low-level image information with high-level prior knowledge, has shown its unique advantage and comprehensive applicability in the segmentation
7、 of medical images especially when apply to complicated and diverse structures. In this paper, by taking of segmenting Musculoskeletal MRI as instance，we propose some improvements to the level-set geometrical active contour model, and validate the validity of the algorithms in the segmentation of no
8、rmal bones and focus. The main work and contributions are as follows: (1) We propose a local adaptive active contour model variational level-set. Compared to the variational level-set without re-initialization which proposed by Li, there are two points of improvement. First, using of the local chara
9、cteristic of image to implement adaptive evolution of the curve, it can improve the segmentation performance when the initial curve crossing the edge of the object; Second, by applying the global C-V energy to the local region and improving the edge stop function，make the model more robust to the ov
10、erlapping of intensity distribution、 intensity inhomogeneity. Strengthen the applicability of the algorithm. Lastly, based on the characteristics of serial medical images, transform the segmentation result of current slice to Piecewise Constant Function using the Heaviside function. Then cast it to
11、the next slice as the initial curve to implement automatic segmentation of serial images. The proposed method has been applied to segment the tumor focus in musculoskeletal MRI. (2) Due to the complicated structure, low contrast and weak edge between different parenchyma, the level-set method usuall
12、y can not obtain a satisfying result when segment cartilage in the knee joint MRI, as it only use low-level image information. In this paper, an integrated active model is proposed by blending prior shape information with the level-set method. Firstly, we use the points distributing model to regulat
13、e the prior sample shape of the object, and then describe the mean shape as the zero-level of the level-set function. Based on the adaptive active model, integrate the shape information into the level-set framework, and use it to guide and restrict the evolvement of the curve. This algorithm can imp
14、rove the identification of weak edge and reduce the interference of non-objects edge. Moreover, it has a good performance of anti-noise. Finally accomplish some elementary researches on the cartilage segmentation of knee joint MRI. Key words: medical image segmentation; active contour model; level-set; local adaptive; prior shape 插图清单 插图清单 图 1-1 常用形变模型分割方法分类.4 图 2-1 三种典型的曲线演化运动.9 图 2-2 水平集函数.10 图 2-3 水平集函数动态演化.10 图 2-4 平均曲率演化.11 图 2-5 GAC 能量表示图 .16 图 2-6 GAC 模型分割结果.
