基于多尺度空间滤波结合两级l_1范数最近邻分类的乳腺微钙化图像病变类型诊断系统  

The Lesion Type Recognition of Mammograms Based on Multi-scale Space Filtering and l_1 Norm Nearest Neighbor Classification

作　　者：胡正平[1] 刘敏华[1] 

机构地区：[1]燕山大学信息科学与工程学院,秦皇岛066004

出　　处：《中国生物医学工程学报》2010年第4期524-530,共7页Chinese Journal of Biomedical Engineering

基　　金：河北省自然科学基金(F2008000891);燕山大学博士基金(B287);中国博士后自然科学基金(20080440124);第二批中国博士后基金特别资助(200902356)

摘　　要：提出一种基于多尺度空间滤波结合l1范数最近邻分类的乳腺微钙化图像病变类型诊断系统,该系统主要由三部分组成:微钙化点检测、微钙化簇特征提取和分类判决。首先利用多尺度空间滤波方法得到原图像的多尺度显著特征图,然后通过基于视觉特性的钙化点分割方法得到粗检测钙化点的二值图像,并送入l1范数最近邻分类器去除假阳性点得到钙化点信息,进而得到局部区域的微钙化簇图像,通过提取钙化簇特征信息,并再次利用l1范数最近邻分类器进行分类,判断该钙化簇的良/恶性。仿真实验结果表明,所设计的微钙化簇检测算法可较好地检测出可疑钙化点,对乳腺图像病变部位进行自动识别,并准确判别病变类型,为医生的诊断提供有效的帮助。Based on multi-scale space filtering and two levle l1 norm nearest neighbor,a lesion type recognition algorithm of mammograms was proposed in this paper.Firstly,the multi-scale salience feature images were obtained using multi-scale space filtering,then the coarse detection binary image of microcalcifications was induced using microcalcifications threshold method based on human visual model,and the false-positive microcalcification points was removed by l1 norm nearest-neighbor（l1 NN） classifier.Finally,the feature information of microcalcification clusters was extracted to determine the microcalcification clusters type（benign or malignant） using the l1 norm nearest-neighbor classifier again.Simulation experiment results demonstrated that the proposed microcalcification clusters detection algorithm could detect the suspicious microcalcifications effectively and identify the lesion automatically from the mammograms.

关 键 词：微钙化点检测 尺度滤波 乳腺图像 病变类型识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]