基于PCA-RF方法的乳腺图像多类别分类研究及应用  

Research and Application of Multiple Classification of Breast Images Based on PCA-RF Method

作　　者：罗姗 邵艳华 LUO Shan;SHAO Yanhua(School of Data Science and Information Engineering,Guizhou Minzu University,Guiyang 550025,China)

机构地区：[1]贵州民族大学数据科学与信息工程学院,贵阳550025

出　　处：《湖北民族大学学报（自然科学版）》2020年第4期402-407,共6页Journal of Hubei Minzu University:Natural Science Edition

基　　金：贵州省科技厅项目(黔科技基础[2017]1085).

摘　　要：由于乳腺图像特征维度大、特征交叉性高及有很高相似性,导致特征分类准确率不高,运行时间长,效率相对低下.提出一种基于PCA-RF进行乳腺图像多类别分类的研究方法,首先对MIAS乳腺图像进行预处理,然后运用PCA算法对乳腺图像提取的特征进行多层次降维,之后使用KNN、AdaBoost、RF对其特征进行分类.分类实验表明,对PCA降维得到乳腺图像特征再进行分类,KNN运行时间效率提高了180倍,RF分类的运行时间也减少了3.5倍.基于PCA-RF方法的乳腺图像正常、良性及恶性分类得到了93.75%分类准确率及95%的敏感性,相对于其他乳腺图像正常、良性及恶性三类别的分类方法得到的结果有所提升.Due to the large feature dimension,high feature intersection and high similarity of breast images,feature classification accuracy is low,running time is long and efficiency is relatively low.This paper proposes a research method of breast image classification based on PA-RF.First,MIAS breast images are preprocessed,then features extracted from breast images are dimensionally reduced on multiple levels using PCA algorithm,and then their features are classified using KNN,AdaBoost and RF.Classification experiments showed that the operating time efficiency of KNN was improved 180 times and the operating time of RF classification was also reduced 3.5 times by classifying breast image features obtained by PCA dimension reduction.The classification accuracy and sensitivity of normal,benign and malignant breast images based on PA-RF method were 93.75%and 95%,which were improved compared with the three classification methods of normal,benign and malignant breast images.

关 键 词：多类别分类 乳腺图像 感兴趣区域 主成分分析 随机森林 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]