一种新的免疫组化图像分割算法研究  

A NOVEL ALGORITHM FOR IMMUNOHISTOCHEMICAL IMAGE SEGMENTATION

作　　者：王慧[1] 江锋[2] 叶永安[2] 李为东[3] 闵乐泉[1,3] 

机构地区：[1]北京科技大学信息工程学院,北京100083 [2]北京中医药大学东直门医院,北京100700 [3]北京科技大学应用科学学院,北京100083

出　　处：《计算机应用与软件》2011年第6期54-56,共3页Computer Applications and Software

基　　金：国家自然科学基金项目(60674059);北京中医药大学自主选题项目(2009JYBZZ-JS078)

摘　　要：免疫组化图像自动分割在病理切片染色分析中有重要的应用价值。对大鼠肝脏免疫组化彩色图像进行深入研究,提取出新的阳性产物图像特征:所有阳性像素满足r>g>b;每个阳性像素在(r-g)、(g-b)、(r-b)三个色差分量上的均值和方差较大。以这些特征为基础,结合中值滤波,提出一种新的阳性产物自动分割算法。利用82幅图像对算法进行验证,与医生目视鉴别结果一致的约占80%。与现有算法的比较结果表明,该算法更适合低放大倍数的大鼠肝脏免疫组化彩色图像的分割,在免疫组化彩色图像的自动分析方面,做了一次有益的探索和尝试。今后还需进一步与病理专家沟通,提高识别率。It＇s very important to automatically segmenting immunohistochemical images in analyzing medical stained biopsies for the diagnosis of some terrible diseases.New image features for positive area are found according to the characteristics of immunohistochemical color images for mouse liver.Firstly,all positive pixels satisfy rgb.Secondly,Red,Green and Blue component image are extracted from RGB color image respectively.rg of color aberration gray-level image is derived from Green component subtracted from Red component.And so do gb and rb.The mathematical expectations and standard deviations of positive pixels on rg,gb and rb are both larger than those of nonpositive pixels.All positive pixels will be extracted based on these features.Finally,median filter algorithm is used to remove the noise produced in the last step.Experiments with 82 images showed that 80% results are in accordance with the estimation of doctors.In comparison with other methods,our algorithm is more appropriate and effective for mouse liver immunohistochemical images with lower magnification.This investigation is very meaningful for quantitative immunohistochemical image analysis.More communication with physicians should be made in future to improve the recognition rate.

关 键 词：图像处理 免疫组化图像分割 特征提取 数学期望 方差 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]