PCA和SVM在新疆哈萨克族食管癌图像分类中的研究与应用  

Classification of Esophageal Cancer Images for Xinjiang Kazak Nationality Based on PCA and SVM

作　　者：杨芳[1] 木拉提.哈米提 严传波[2] 姚娟[3] 阿布都艾尼.库吐鲁克 孙静[2] Yang Fang;Hamit Murat;Yan Chuanbo;Yao Juan;Kutluk Abdugheni;Sun Jing(College of Medical Engineering Technology,Xinjiang Medical University,Urumqi 830011,China;Department of Medical Engineering,the Affiliated Tumor Hospital,Xinjiang Medical University,Urumqi 830011,China;Department of Radiology,the First Affiliated Hospital,Xinjiang Medical University,Urumqi 830054,China)

机构地区：[1]新疆医科大学附属肿瘤医院医学工程科,乌鲁木齐830011 [2]新疆医科大学医学工程技术学院,乌鲁木齐830011 [3]新疆医科大学第一附属医院影像中心,乌鲁木齐830054

出　　处：《科技通报》2017年第2期43-47,108,共6页Bulletin of Science and Technology

基　　金：国家自然科学基金(81460281,81560294,81160182,61201125)

摘　　要：目的:利用PCA和SVM对新疆哈萨克族食管癌X射线图像进行特征提取、特征选择及分类研究。方法:利用基于灰度共生矩阵的纹理特征和小波变换的频域特征提取法,提出将ROC曲线面积选择法和主成分分析法相结合的两步式特征选择法,利用Bayes和SVM分类器对图像进行分类以验证所提取特征的分类能力。结果:AUC>0.7的特征经主成分分析后输入到SVM分类器和Bayes分类器中得到的分类准确率和AUC值最高,分别为91%和85%、0.945和0.915。结论:SVM具有较好的分类性能,两步式特征选择法能有效地消除特征之间的共线性,极大提高了特征的分类能力,本研究有望提高新疆哈萨克族食管癌CAD系统的整体性能。Objective:This paper details the feature extraction, feature selection and classification of X-ray images for Xinjiang kazak esophageal cancer based on PCA and SVM. Methods:Texture features and frequency features are extracted by gray level co-occurrence matrix and wavelet transformation method.A two-stage feature selection process is used to select the features with high classification ability. The receiver operating characteristic (ROC) curves of each feature are calculated to find the features with higher classification accuracy. Features with areas under ROC curve (AUC) bigger than 0.7 are adopted.Then, principal component analysis (PCA) is applied to the set of image features selected from stage one.Both SVM and Bayes classifiers are employed to classify the images by type. Results:Experimental results show that the highest classification performance is achieved when the features selected by AUC>0.7 and PCA are employed. The accuracy and AUC value are 91% and 85%, 0.945 and 0.915 for SVM and Bayes classifier, respectively. Conclusions:Classification performance of SVM outperforms the Bayes.The two- stage feature selection method can improve the classification performance significantly by removing redundancy due to highly correlated features. Therefore, the proposed method is promising for the diagnostic of esophageal cancer in Xinjiang Uygur autonomous region.

关 键 词：新疆哈萨克族食管癌 特征提取 主成分分析 支持向量机 

分 类 号：R730.44[医药卫生—肿瘤]