基于级联分类器的乳腺癌病理学图像中有丝分裂检测  

Mitosis detection in breast cancer histopathology with cascaded classifier algorithm

作　　者：唐娇[1] 梁毅雄[1] 邹北骥[1] 周其亚 Tang Jiao Liang Yixong Zou Beiji Zhou Qiya(Institute of Information Science ＆ Engineering, Central South of University, Changsha 410083, China)

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《计算机应用研究》2016年第12期3876-3879,3888,共5页Application Research of Computers

基　　金：湖南省自然科学基金资助项目(14JJ2008);国家自然科学基金资助项目(61573380)

摘　　要：有丝分裂检测与计数是诊断乳腺癌的一个重要指标。针对乳腺癌病理图像中的有丝分裂核与非有丝分裂核难区分、难检测等研究难点,提出了一个基于级联分类器的计算机辅助有丝分裂检测算法。考虑到分割后正负样本在数量上的差异性较大,该方法通过一个级联分类器在其每一层中丢弃一部分非有丝分裂块,使得级联中后一个分类器在训练时能更关注那些比较难分的样本;利用颜色直方图找出有丝分裂块与非有丝分裂块差异性较大的颜色通道,以便提取的特征具有更好的分类性能。该方法最终以F-measure值为0.732验证了所提出算法的有效性。Mitotic count is an important parameter in the diagnosis of breast cancer. For the difficulties in mitotic detection, this paper proposed a computer-aided mitosis detection algorithm based on a cascaded classifier. By the consideration of the large difference between the number of mitosis and non-mitosis, this paper proposed a method for combining weak classifiers in a cascade which allowed part of non-mitosis regions of the image to be quickly discarded while spending more computation on promising mitosis-like regions. It could find several color channels who had the best recognition performance in mitosis and nonmitosis by color histogram, so that the features extracted in them had good classification. This method can achieve F-measure 0. 732 eventually, the result of experiment, demonstrates the effectiveness of the proposed algorithm.

关 键 词：级联分类器 乳腺癌 病理图 有丝分裂检测 颜色直方图 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]