基于着色分离的免疫组化图像核分割研究

摘要

Nuclei segmentation is a crucial step in immunohistochemical (IHC) quantitative analysis. We present a nuclei segmentation technique based on stain separation in the paper. First, the colour deconvolution is used to conduct stain separation on multi-stained IHC image, in turn the improved SCFCM algorithm is utilised to make coarse separation on single stained greyscale image, and then the watershed algorithm is employed to separate the clustered nuclei, at last the post-processing is done through nucleus size analysis to fulfil the precise nuclei segmentation of haematoxylin stained or multi-stained IHC image. In experiments we adopt nine breast cancer sample images with about 1000 nuclei as our data. Compared with manual sketched results, we derive that the detection rate of nuclei is 92. 66%. Experiment demonstrates that the technique presented in this paper has high nuclei segmentation accuracy on IHC image with pretty good robustness.%细胞核的分割是免疫组化定量分析中非常关键的一步.提出一个基于着色分离的核分割方法,首先采用颜色去卷积算法对多着色的IHC (immunohistochemical)图像进行着色分离,进而利用改进的SCFCM算法对单着色灰度图像进行粗分割；然后利用分水岭算法分离粘连细胞；最后通过细胞核尺寸分析进行后处理,完成对苏木素或多种染色的免疫组化图像的准确核分割.实验采用9幅乳腺癌样本图像,约1000个细胞核作为数据.与手工勾画结果进行对比分析,得出细胞核检测率为92.66％.实验表明,该方法对免疫组化图像核分割准确率高,且具有较好的鲁棒性.