基于支持向量机和椭圆拟合的细胞图像自动分割

发布时间：2019-04-11 15:32

【摘要】：提出细胞显微图像的自动分割方法,重点解决重叠细胞之间的分割难题.对原始图像进行二值化,提取每个连通区域的多个形状特征并应用支持向量机(SVM)进行测试,区分单个与重叠细胞.运用瓶颈检测寻找分离点对分割重叠细胞.基于细胞的椭圆或圆形结构,对分割后的边缘,应用改进的椭圆拟合法进行修正,修正后的细胞区域能够有效地反映重叠在一起的细胞的真实形状.对新生成的封闭区域进行循环检测,直至所有重叠细胞分割完毕.实验结果表明,相较于多种现有的细胞分割方法,采用该算法能够更有效地抑制过分割和欠分割,分割准确率、敏感度和特异度高.
[Abstract]:The method of automatic segmentation of the microimages of the cells is proposed, and the problem of the segmentation between the overlapping cells is mainly solved. The original image is binarized, a plurality of shape features of each connected region are extracted, and a support vector machine (SVM) is applied for testing to distinguish single and overlapping cells. The method of bottleneck detection is used to find the separation point pair to separate the overlapping cells. Based on the elliptical or circular structure of the cell, the modified elliptical quasi-legal modification is applied to the segmented edge, and the modified cell region can effectively reflect the true shape of the cells that overlap together. The newly generated closed region is subjected to cyclic detection until all overlapping cells are segmented. The experimental results show that the method can more effectively inhibit the over-segmentation and under-segmentation, the segmentation accuracy, the sensitivity and the specificity compared with a variety of existing cell segmentation methods.
【作者单位】：中南大学信息科学与工程学院;湖南科技大学计算机科学与工程学院;中南大学生物医学工程研究所;胜利石油管理局滨南医院;
【基金】：国家自然科学基金资助项目(61379107,61172184,61174210,61402539) 教育部“新世纪优秀人才支持计划”资助项目(NCET-13-0603) 高等学校博士学科点专项科研基金资助项目(20130162110016) 湖南省科技基本建设项目(20131199) 湖南省科技计划资助项目(2015RS4008) 湖南省自然科学基金资助项目(2015JJ4061,2017JJ3091) 中南大学中央高校基本科研业务费专项资金资助项目(2014ZZTS053) 湖南省研究生科研创新资助项目(CX2014B052)
【分类号】：TP391.41;TP18

【相似文献】