灰度形态学运算对消除分水岭过分割的影响研究  

Influence of Gray-scale Morphological Operations on Eliminating Over-segmentation of Watershed Segmentation

作　　者：李志超[1] 王健[1] 

机构地区：[1]第三军医大学附属西南医院放射科,重庆400038

出　　处：《医疗卫生装备》2010年第11期12-16,共5页Chinese Medical Equipment Journal

基　　金：重庆市科技攻关计划项目(CSTC;2008AB5038)

摘　　要：目的:运用分水岭算法对图像进行分割存在着过分割的现象,需要结合一定的预处理方法来减少这种影响。就胸部CR图像片段运用灰度形态学运算进行预处理,以研究其对分水岭算法分割结果的影响。方法:选取PACS存储的胸部CR图像20例,对每例病例图像截取5个没有骨骼与异物干扰仅含肺组织结构的图像片段,共计100个小图片。将每个小图片及对其应用碟形结构子进行形态学开运算后的处理图像,分别运用Sobel算子进行卷积得到对应的梯度图像,并对其进行3次分水岭运算。统计原始图片得到了封闭区域数量及预处理后的封闭区域数量、原始图像漏检的结节数量以及预处理后的漏检数量,并统计数据进行配对t检验。结果:原始图像检出区域平均为14.5,预处理后图像检出区域平均为9.48,两者差异相关性为0.656,显著性<0.000 1,t值为11.967,95%置信区间为4.188～5.852,两者具有显著性差异。而两者的漏检区域平均值分别为0.26和0.15,两者差异相关性为-0.021,显著性为0.834,检验t值为1.054,而95%置信区间为-0.097～0.317。可以认为两者没有显著性差异。结论:运用灰度形态学开运算对图片进行预处理可以有效地降低分水岭算法过分割出的假阳性病灶,同时不会增加运用分水岭算法对图像分割的漏检率。Objective To combine watershed operation with some pretreatment methods to reduce the effect of over-segmentation on image,and to apply grayscale morphological opening to pretreatment of image fragments of CR chest to study the influences on the results of segmentation in Watershed Operation.Methods A total of 100 image fragments were cut out from 20 CR chest images stored in PACS.5 image fragments of each image only contained lung issues without skeletons and foreign-body interferences.Every small image and processed image after the operations by disk structure were convoluted with Sobel Edge operation and corresponding images were obtained.Original images were counted to obtain the number of enclosed shapes,post-pretreative enclosed shapes,segments undetected by original images,and post-pretreative undetected segments.These statistics were performed with paired t-test.Results The mean of regions detected by original images was 14.5,while that detected by post-pretreated images was 9.48（r=0.656,p0.0001,t=11.967）.The 95% confidence interval was 4.188-5.852,which showed there were no significant differences between the two regions.The means of undetected regions were 0.26 and 0.15,respectively（r=-0.021,p=0.834,t=1.054）.The 95% confidence interval was-0.097~0.317,which suggested that there were no significant differences between these two undetected regions.Conclusion Image pretreatment with Grayscales Morphological Opening can effectively reduce false-positive lesions over-segmented with Watershed Operation,and not increase undetected rate of segmentation with it.

关 键 词：图像处理 计算机辅助 灰度形态学 分水岭 过分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]