医学图像分析中的分割算法研究
文献类型：学位论文
| 作者 | 朱万琳 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2006-05-28 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 蒋田仔 |
| 关键词 | 图像分割 变分 J散度 模糊C均值 核磁共振图像 正电子断层扫描图像 蛋白质凝胶电泳图像 Image segmentation variational J Divergence FCM MRI PET Electrophoresis |
| 其他题名 | Segmentation in Medical Image Analysis |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 医学图像分割是所有医学图像分析的基础，也是基于医学影像进行人体解剖与功能分析的基础。它是指向性的，对于不同的模态图像，不同的解剖结构，需要不同的分割方法。它又是一个病态的问题，因为图像的伪影以及人体器官的复杂，使得分割问题不存在唯一解。随着医学图像成为日常治疗与研究的主要手段，以及医学图像分析技术的快速发展，都需要更加准确，可靠，鲁棒的分割算法。本文从综述现有的分割方法学开始，给出我们对分割算法的一些尝试，论文的主要贡献如下： 我们提出一种新的变分分割框架。这个方法的创新性在于用J散度(对称的K-L散度)来度量局部区域与全局区域的不相似性。每个像素的灰度值被其邻域的概率分布代替。它的引入确保了分割算法对噪声的鲁棒性。J散度被用来度量局部和全局分布的概率密度函数之间的``距离"，所有的局部与全局分布都被假设成高斯分布，这个假设能够使计算效率得到极大的提高。此外基于边缘的测地线能量也被加入到提出的能量泛函中，这样做能够减少由区域项能量在物体边缘处带来的分割误差。而且提出来的方法是多相位的。从合成试验到真实数据的试验证明了我们方法的有效性。 我们把提出来的J散度变分分割框架应用到核磁共振图像(MRI)脑组织分割问题上。由于脑组织可以被分为白质，灰质和脑脊液，它们之间是两两相交的，我们实现一种三相位的分割模型，与其它方法相比，它的优点是能够保证任意两个区域都形成竞争。从而能够保证对初始误差更加鲁棒。 我们也提出一种迭代的模糊C均值(FCM)方法去分割正电子断层扫描图像(PET)中的肿瘤。FCM方法是基于直方图的快速算法。用传统的FCM必须指定分类数目，然而这种背景污染的图像其分类数目对于不同的图像可能不同。在我们的方法中我们只指定两类：前景和背景，计算前景的标准差来决定是否需要进行迭代，这个方法的优点是完全自动而且简单，我们也把这个方法应用到蛋白质凝胶电泳图像中的蛋白质点检测上，所有的试验结果证实了提出的方法的有效性和效率。 |
| 英文摘要 | Image segmentation that partitions an image into non-overlapping meaningful regions plays key role in medical image analysis. It is the basis of further structure and functional analysis. It is specific for different modality images and different anatomy structures, a ``unified" method never exists as a result. It is ill-posed due to the image artifacts and anatomy structures complexity of the human body. With the acquisition of medical image has become a routine task for clinical and research application, also the rapid development of medical image analysis technique, more accurate, reliable and robust segmentation methods are required. Starting from an overview of segmentation methodology, we present our proposed segmentation methods and contributions that summarized as follows: A novel variational segmentation framework is proposed. The originality of our approach is on the use of J-divergence (symmetrized Kullback-Leibler divergence) to measure the dissimilarity between local and global regions. The voxel intensity value is replaced with probability distribution of its neighborhood. It is introduced to ensure the robustness of the algorithm when image corrupted by noise. Then, J-Divergence is used to measure the ``distance" between local and global region probability density function. The distribution of all local regions and global regions are assumed to follow Gaussian distribution. The assumption brings the computation efficiency. In addition, a geodesic based energy constraint is imposed on the local and global region based energy functional to decrease segmentation errors in the vicinity of object boundary. The proposed method is designed for any-phase segmentation problems. From the synthetic and real brain MR images experiments, they proved the effectiveness of the proposed method. We also propose an automatic iterative fuzzy C means (FCM) to delineate brain tumors from PET images. The FCM is based on image histogram so it is far efficient than conventional FCM. When using conventional FCM, in order to extract tumor from a PET image, we have to specify the numbers of clusters and which may vary from one to another image. In the proposed method, the whole image is separated into two parts: background and foreground. The standard deviation of foreground is calculated to decide whether the iteration should stop. The advantage of the algorithm is completely automatic and simple. The method is also performed for 2D electrophoresis images. All the experimental results validate the effectiveness and efficiency of this approach. |
| 语种 | 中文 |
| 其他标识符 | 200218014603239 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/5917]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 朱万琳. 医学图像分析中的分割算法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2006. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。