摘要
针对细胞图像尺寸大、细胞形状各异,导致从图像中分割出精准的细胞十分困难的问题,以卷积神经网络为基础,结合染色校正方法和简单线性迭代的超像素聚类算法,提出了一种新的结构来进行细胞图像分割。利用染色校正方法对细胞图像进行预处理,提高图像的颜色对比度;利用卷积神经网络获得初步分割结果;将简单线性迭代聚类获得的超像素边界信息反馈到初分割图像上进行改进提升。提出的算法可以有效地减少图像局部信息的冗余,更准确地获得目标区域的边界位置。实验表明,提出的算法细胞分割准确率达到了92.72%,与经典卷积神经网络、阈值分割等其他细胞分割算法相比,具有更好的分割效果。
Cell images have the characteristics of large sizes and different cell shapes.Therefore,it is very difficult to segment the precise cells from the medical images.Based on the convolution neural network,combining with the stain regularization method and the simple linear iterative clustering algorithm,this paper proposed a new structure for cell segmentation.First of all,stain regularization method could improve the color contrast of the images in preprocessing.Then the convolution neural network could get the initial cell segmentation result.At the same time,using super pixel clustering method could obtain the boundary information of cell images.Finally the boundary information was feedback to the initial cell segmentation result to improve the accuracy of the segmentation.The proposed algorithm could effectively reduce the redundant local information in image,and got the boundary of the target cell area more accurately.Experimental result shows that the segmentation accuracy of the proposed algorithm is 92.72%.Compared with the classical convolution neural network and other cell segmentation methods such as the threshold segmentation method,the proposed algorithm has better segmentation results.
作者
杨金鑫
杨辉华
李灵巧
潘细朋
刘振丙
周洁茜
Yang Jinxin;Yang Huihua;Li Lingqiao;Pan Xipeng;Liu Zhenbing;Zhou Jieqian(School of Automation,Beijing University of Posts&Telecommunications,Beijing 100876,China;School of Computer Science&Information Security,Guilin University of Electronic Technology,Guilin Guangxi 541004,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第5期1569-1572,1577,共5页
Application Research of Computers
基金
国家自然科学基金(21365008
61562013)
广西自然科学基金项目(2013GXNSFBA019279)
广西重点研发计划项目(2016AB14005)
关键词
细胞分割
卷积神经网络
超像素聚类
染色校正
乳腺细胞图像
cell segmentation
convolution neural network
super pixel clustering
stain regularization
breast cell image
