异常宫颈细胞核的自适应局部分割  

Segmentation of abnormal cervical nuclei using an adaptive and local approach

作　　者：张灵[1] 李静立[2] 陈思平[1,2,3] 汪天富[3] 江少锋[2] 刘少雄[4] 

机构地区：[1]浙江大学生物医学工程与仪器科学学院,杭州310027 [2]南昌航空大学测试与光电工程学院,南昌330063 [3]深圳大学医学院广东省生物医学信息检测和超声成像重点实验室,深圳518060 [4]深圳市南山区人民医院病理科,深圳518052

出　　处：《中国图象图形学报》2013年第10期1329-1335,共7页Journal of Image and Graphics

基　　金：国家自然科学基金重点项目(61031003);国家自然科学基金项目(60772147;60871060)

摘　　要：为实现宫颈液基细胞图像中异常细胞核的准确分割,提出一种新的自适应局部细胞核分割方法。在自适应阶段,采用一种利用灰度和纹理信息的快速自适应阈值算法大致检测出细胞核区域;在局部阶段,对每一个粗分割得到的连通区域,在其局部邻域内,使用一种利用边界和区域信息的、基于泊松概率分布的图割法修正分割结果。将此方法用于苏木素&伊红染色的宫颈液基细胞图像,结果显示,本文方法的平均计算时间为1.6 s/幅,且比2012年Li等人提出的宫颈细胞分割方法在细胞核检测率、和异常细胞核分割精度上均提高了19.7%。Abstract： For accurate segmentation of abnormal nuclei in liquid-based cervical cell images, a new nuclei segmentation method is proposed, which uses adaptive and local strategies. The adaptive stage detects each nucleus region approximately by applying an efficient adaptive thresholding algorithm that uses intensity and texture information. The local stage refines each coarse segment within its local neighborhood by using a Poisson distribution based graph cuts, which utilizes boundal~~ and region information. The proposed method is applied to Hematoxylin ＆ Eosin stained liquid-based cervical cell images. The results show that the proposed method achieves a speed of 1.6 s per image, and significantly outperforms a state-of-the- art method by Li et al in 2012 in terms of nuclei detection rate and abnormal nuclei segmentation accuracy, both with a 19.7% improvement.

关 键 词：宫颈液基细胞 图像 H&E染色 异常细胞核 分割 自适应局部分割 图割法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]