中国组织工程研究

中国组织工程研究 ›› 2010, Vol. 14 ›› Issue (35): 6559-6562.doi: 10.3969/j.issn.1673-8225.2010.35.025

• 骨与关节图像与影像 bone and joint imaging • 上一篇    下一篇

Gibbs场先验参数估计在医学图像分割中的应用

李  彬1，刘  同2，陈武凡1   

  1. 1南方医科大学生物医学工程学院，广东省广州市   510515；2 黄河水利委员会信息中心，河南省郑州市  450004
  • 出版日期:2010-08-27 发布日期:2010-08-27
  • 作者简介:李彬☆，男，1964年生，河南省焦作市人，汉族，2007年南方医科大学毕业，博士，副教授，主要从事图像处理和模式识别研究。 libin371@fimmu.com

Estimation of Gibbs field prior parameter applied in medical image segmentation 

Li Bin1, Liu Tong2, Chen Wu-fan1   

  1. 1 School of Biomedical Engineering, Southern Medical University, Guangzhou  510515, Guangdong Province, China; 2 Information Center of Yellow River Conservancy Commission, Zhengzhou  450004, Henan Province, China
  • Online:2010-08-27 Published:2010-08-27
  • About author:Li Bin☆, Doctor, Associate professor, School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, Guangdong Province, China libin371@fimmu.com

PDF

288

摘要：

背景：基于马尔科夫随机场的图像分割算法已经成为医学图像分割的重要方法，其中，Gibbs场先验参数的取值对分割精度有很大的影响。
目的：根据脑部MR图像的成像特点，探讨Gibbs场先验参数的估计方法，从而提高图像分割的精度。
方法：通过对脑部MR图像的统计分析，得到图像高斯噪声的方差与Gibbs场先验参数的对应关系。然后在基于马尔可夫随机场图像分割算法的迭代过程中，根据高斯分布的方差估计值，用插值方法估计Gibbs场先验参数。
结果与结论：通过对模拟脑部MR图像和临床脑部MR图像分割实验，表明该方法比传统的设定Gibbs场先验参数为某一常数的方法有更精确的图像分割能力，并且实现了图像的自适应分割，具有方法简单、运算速度快、稳健性好的特点。

关键词: 医学图像分割, 马尔可夫随机场, 参数估计, Gibbs场, 先验参数

Abstract:

BACKGROUND: The method based on Markov random field model is an important segmentation algorithm for medical images. The prior parameters of Gibbs field can severely affect the accuracy of image segmentation.
OBJECTIVE: Based on the properties of medical images, to explore the estimation method of Gibbs field prior parameter to improve the accuracy of segmentation.
METHODS: The relation between the variance of Gauss noises and the optimal Gibbs field prior parameters for brain MR images was obtained by a statistical method. In the image segmentation iterative steps, the Gibbs field prior parameters were estimated by means of interpolation using the estimation of image variance.
RESULTS AND CONCLUSION: Simulated brain MR images with different noise levels and real brain MR images are presented in the experiments. The results show that the proposed estimation method is easy in practical implementation, faster in computational speed, and is capable of achieving finer and adaptive segmentation compared with conventional methods.

中图分类号: