中山大学学报自然科学版 ›› 2010, Vol. 49 ›› Issue (2): 53-59.

• 研究论文 • 上一篇    下一篇

基于图的脑组织磁共振图像分割方法

张竞丹1,2   

  1. (1.中山大学数学与计算科学学院,广东 广州 510275;2.深圳信息职业技术学院 电子通信技术系,广东 深圳 518029)
  • 收稿日期:2009-04-22 修回日期:1900-01-01 出版日期:2010-03-25 发布日期:2010-03-25

GraphBased Hierarchical Clustering Method for Automatic Brain MR Image Segmentation 

ZHANG Jingdan1,2   

  1. (1.Department of Mathematics, Sun Yatsen University, Guangzhou 510275, China;2. Department of Electronic Communication Technology, Shenzhen Institute of Information Technology, Shenzhen 518029, China)
  • Received:2009-04-22 Revised:1900-01-01 Online:2010-03-25 Published:2010-03-25

摘要: 提出一种基于图的层次聚类算法实现脑组织磁共振图像的自动分割。首先,采用基于图的分割方法对脑组织MR图像进行初始分割。由于脑组织MR图像各类组织结构分布复杂,尤其是脑脊液和灰质区域细节信息丰富、结构变化多样,分割结果中存在过分割现象。因此,利用对偶树复小波变换高频子带信息构造基于图的分割方法中参数k的自适应取值函数,避免图像平滑区域分割后产生大量小区域。然后,以层次聚类算法合并分割得到的小区域,解决基于图的方法分割脑组织MR图像中存在的过分割问题。最后,通过大量真实脑组织MR图像实验证明该方法在脑组织MR图像分割中的准确性和稳定性。

关键词: 基于图的算法, 对偶树复小波变换, 图像分割, 脑组织MR图像, 层次聚类算法

Abstract: A graphbased hierarchical clustering (GBHC) method for brain MR image segmentation is presented. Firstly, the standard graphbased method is applied to produce a coarse segmentation of brain MR image. However, the segmentation result of the graphbased method is oversegmentation because of the complicated structure of brain. So, we apply an adaptive function to control the value of parameter kin the graphbased method, which integrates the information from the highfrequency subbands of dualtree complex wavelet transform. Then, the hierarchical clustering method is used to merge the oversegmented regions in the segmentation result. The method is validated by extensive experiments using real T1weighted MR images, and compared with the stateoftheart algorithms.

Key words: graphbased method, dualtree complex wavelet transform, image segmentation, brain MR image, hierarchical clustering method

中图分类号: