KNN分类器在新疆哈萨克族食管癌分型中的应用  

Classification on Xinjiang Kazak Esophageal Disease Based on KNN Classifier

作　　者：张岁霞 严传波[2] 木拉提·哈米提[2] 阿布都艾尼·库吐鲁克[2] 孙静[2] 艾赛提·买提木沙[3] 姚娟[4] 杨芳[1] 伊力扎提·阿力甫 孔喜梅[1] 

机构地区：[1]新疆医科大学基础医学院,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]新疆医科大学公共卫生学院,乌鲁木齐830011 [4]新疆医科大学第-附属医院,乌鲁木齐830011

出　　处：《科技通报》2016年第8期46-50,共5页Bulletin of Science and Technology

基　　金：国家自然科学基金(81160182;81460281;61201125);江西民族传统药协同创新项目(JXXT201401001-2);留学人员科技活动择优项目(2013)277号

摘　　要：目的:探讨KNN(k-Nearest Neighbor algorithm,K近邻结点算法)分类器在新疆哈萨克族食管癌分型中的应用。方法:采用KNN分类器,依据食管癌的灰度-梯度共生矩阵和灰度共生矩阵特征值,对其分型。选取样本量的40%、50%、60%作为三个训练集,K取1-29,训练并得到最优K值;选取样本量的10%到100%(以10%递增率)10个测试集,验证结果;获得最佳KNN分类模型,对模型进行评估。结果:训练结果:当K=1时,三种食管癌都能获得最高分类准确率。测试结果:改变测试集大小,当验证数据量增大时,分类准确率随之增加。最佳KNN分类模型评估:该KNN分类模型有一定的准确度,可以得到可靠的分类结果。结论:KNN分类器为新疆哈萨克族食管癌分型提供一定的依据,也为新疆哈萨克族食管癌的计算机辅助诊断系统的研发奠定基础。Objective：This work shows a detailed description of the KNN（k-Nearest Neighbor algorithm）classifier when using it for Kazakh esophageal cancer classification. Methods：Using MATLB topreprocess and extract features based on gray gradient co- occurrence matrix and gray level cooccurrencematrix （GLCM） texture features,Using KNN classifier Classification on image features. First,selecting 40%、50%、60% of total sample data as training sets, We set the number of neighbors k from 1to 29,training and getting optimal k value;Second,using ten different sizes as the classification set,whichset from 10% to 90% of data（in increments of 10%） ,validating the optimal k value; Third,evaluating thebest KNN classification model. Results Training Results： the optimal k value is K=1;ClassificationResults： accuracy with the increasing of the percentage of the data used for Classification increasing Model evaluation： KNN classification model has a certain degree of precision and reliable.Conclusion：KNN classifier can improve the classification ability and provide a certain basis judgment of Kazakh esophageal cancer classification,this laid the foundation of computer diagnosis system for Kazakh esophageal cancer.

关 键 词：KNN算法 灰度-梯度共生矩阵 灰度共生矩阵 图像分类 

分 类 号：R735.1[医药卫生—肿瘤] TP751.1[医药卫生—临床医学]