基于人工交互分水岭区域合并的医学图像分割研究  

A Study of Medical Image Segmentation Based on Human Interactive Region Merges of Watershed

作　　者：吕晓琪[1] 范运洲[1,2] 谷宇[1] 刘静[1] 

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010 [2]上海青蓝科技有限责任公司,上海200030

出　　处：《中国医学影像学杂志》2010年第6期516-520,共5页Chinese Journal of Medical Imaging

基　　金：国家自然科学基金(60761002):基于非刚性配准的多模医学图像融合技术研究

摘　　要：目的:解决分水岭算法存在的过分割[1]问题,对分水岭算法的合并方法进行改进。材料与方法:主要思想是在前两次合并的基础上进行第三次局部合并,用人机交互[2]的方式留下感兴趣的部分,其余部分继续合并。结果:这种改进的方法使过分割问题得到了很好的解决,而且医学图像中的病变小区被保留下来。结论:此人工交互的方法效果优于传统的分水岭方法。Purpose To solve the problems of over-segmentation and improve the merger method in the watershed algorithm.Materials and Methods The main measure was to implement the third partial merger based on the former two mergers and to keep the region of interest using human-machine interactive methods.Results Using the improved method,the problems of over-segmentation were well restrained.In addition,the small regions with pathological changes were well segmented.Conclusions The method of human interaction is better than the traditional watershed approach.

关 键 词：人机交互 分水岭算法 过分割 图像分割 

分 类 号：R814.4[医药卫生—影像医学与核医学]