基于图割和K聚类的CT图像分割

DSCT cardiac image segmentation based on K means and graph cuts

目的 :结合K聚类方法和图割理论完成CT心脏图像的自动分割,提取出完整、精确的心脏结构。方法 :第一步运用各向异性扩散去噪,去除图像噪声;第二步采用K聚类方法确定像素初始标号,然后在此基础上建立关于标号的能量函数,构造网络图;最后运用最大流/最小割算法实现能量函数最小化,完成对感兴趣区域的分割。结果:实现了双源CT(dual source CT,DSCT)心脏图像感兴趣区域的分割,取得了较好的结果。结论:利用图割和K聚类算法相结合可以实现DSCT图像心脏结构的快速、鲁棒、精确分割。

Objective To propose a mixed segmentation method that combines clustering with graph cuts to automatically segment the dual-source CT heart image and further accurately extract the cardiac structure. Methods Firstly, the noise of image was filtered by anisotropic diffusion. Then the preliminary segmentation was performed after K clustering was used to determine optimal threshold. Graph cuts of energy minimization was .applied to extracting the contour of interested region. Then the segmentation result was obtained. Results Region of interest was segmented for DSCT cardiac image. ConclosionExperimental results show that the combination of graph cuts and K clustering algorithm can robustly and precisely segment DSCT image cardiac structure.