中国组织工程研究与临床康复 第 15 卷 第 39 期 2011–09–24 出版
Journal of Clinical Rehabilitative Tissue Engineering Research September 24, 2011 Vol.15, No.39
ISSN 1673-8225 CN 21-1539/R CODEN: ZLKHAH
Department of
Biomedical
Engineering, School
of Bioscience and
Bioengineering,
South China
University of
Technology,
Guangzhou
510006, Guangdong
Province, China
Ji Jian-peng,
Department of
Biomedical
Engineering, School
of Bioscience and
Bioengineering,
South China
University of
Technology,
Guangzhou
510006, Guangdong
Province, China
163.com
Correspondence to:
Yang Rong-qian,
Doctor, Lecturer,
Department of
Biomedical
Engineering, School
of Bioscience and
Bioengineering,
South China
University of
Technology,
Guangzhou
510006, Guangdong
Province, China
bmeyrq@gmail.com
Supported by:
Central University
Special Fund for
Basic Research
Projects of South
China University of
Technology, No.
2009ZM0235*
Received: 2011 -06-15
Accepted: 2011 -07-13
基于分水岭算法的交互式三维分割方法
*
纪建鹏,黎丽华,杨荣骞,吴效明
Interactive three-dimensional segmentation based on watershed algorithm
Ji Jian-peng, Li Li-hua, Yang Rong-qian, Wu Xiao-ming
Abstract
BACKGROUND: The accurate three-dimensional (3D) segmentation of human tissue can improve the accuracy of clinical
diagnosis, the traditional gradient watershed segmentation has the over-segmentation problem which makes the 3D
segmentation of human tissue hard to implement.
OBJECTIVE: In order to segment human tissue’s 3D structure and reduce the influence of the pseudo-minima points on image
segmentation, to propose an interactive 3D segmentation method based on marker-controlled watershed algorithm.
METHODS: The internal and external markers of CT sequence images were extracted, then the gradient images were modified
to segment the images. According to the similarity of neighbouring slices, we segmented the 3D structure of human tissue using
human-computer interactive method. We first selected a point from the interested area in the fi rst slice. Then we could extract the
interested area from the 3D slices by using the overlap relationship of tissues on continuous CT slices.
RESULTS AND CONCLUSION: The marker-controlled watershed algorithm solves the over-segmentation problem which
segmenting the original gradient images meets. The proposed method makes it possible to make the further segmentation. The
results suggest that, the segmented result using interactive 3D segmentation based on watershed algorithm can clearly and
accurately shows the 3D features of the tissue.
Ji JP, Li LH, Yang RQ, Wu XM. Interactive three-dimensional segmentation based on watershed algorithm.Zhongguo Zuzhi
Gongcheng Yanjiu yu Linchuang Kangfu. 2011;15(39): 7351-7354. [http://www.crter.cn http://en.zglckf.com]
摘要
背景:在临床中准确对人体组织进行三维分割能提高临床诊断的准确性,但传统的分水岭算法存在过度分割问题,难以实
现人体组织的三维分割。
目的:为准确三维分割人体组织,减少图像中伪极小值点对图像分割的影响,提出了一种基于控制标记符分水岭的交互式
三维分割方法。
方法:提取 CT 序列图像的内部和外部标记符,以此修正梯度图像并进行分割;在此基础上,根据序列图像上下层的相似性,
利用人机交互进行组织结构的三维分割。首先在第一张序列图像上手工选取感兴趣区域上的一个点,借助同一组织在连续
CT 序列图像上面积的重叠关系即可从三维序列图上提取出感兴趣区域。
结果与结论:基于控制标记符的分水岭算法解决了直接应用梯度图像进行分割的过度分割问题,便于进一步分割图像。利
用基于分水岭算法的交互式三维分割方法得到的三维分割结果经过三维可视化后可清晰、准确地反映组织的三维特征。
关键词:医学图像;控制标记符分水岭;人机交互;三维分割;数字化医学
doi:10.3969/j.issn.1673-8225.2011.39.033
纪建鹏,黎丽华,杨荣骞,吴效明. 基于分水岭算法的交互式三维分割方法[J].中国组织工程研究与临床康复,2011,
15(39):7351-7354. [http://www.crter.org http://cn.zglckf.com]
0 引言
图像分割是图像处理过程中至关重要的
一部分,在医学影像分析领域,图像分割可以
将医学图像中不同的生物组织分割成不同的
区域
[1]
。现有的图像分割方法主要有基于区域
的分割方法、基于边缘的图像分割方法、模糊C
均值聚类法、分水岭分割算法等
[2-3]
。阈值法是
基于区域的分割方法,是一种简单有效的图像
分割方法,特别是不同物体或结构之间有很大
的强度对比时,能够得到很好的效果。它用一
个或几个阈值将图像的灰度级分为几个部分,
它认为具有同一灰度值的象素属于同一个物
体
[4]
。它的主要局限是,最简单形式的阈值法
只能产生二值图像来区分两个不同的类;另外
它只考虑象素本身的值,一般不考虑图像的空
间特性,这样就对噪声非常敏感
[5]
。区域生长方
法是另一种基于区域的分割方法,它根据图像
中感兴趣区域的性质,如灰度、纹理、色彩等,
把感兴趣区域分割出来;其一般过程为,从预
先选定的生长点向邻域搜寻与生长点灰度差
小于某设定值的点;其优点是操作简单,分割
速度快;缺点是在图像的灰度发生突变或者有
噪声的地方可能会出现小洞,因此需要一定的
后处理
[6]
。
传统的分水岭通过模拟水盆浸盖过程,将
不同水域作为分割图像的一部分
[7]
。该算法对噪
声非常敏感,将导致严重的过分割,图像分割
实效性较差
[8]
。而基于控制标记符的分水岭算法
万方数据