数据挖掘算法对乳腺肿瘤超声图像特征的优化及良恶性分类研究  

Optimization and Classification of Ultrasound Image Features for Breast Tumor Based on Data Mining Algorithm

作　　者：孙梅[1] 严传波[1] 张雨[2] 毕雪华[1] Sun Mei;Yan Chuanbo;Zhang Yu;Bi Xuehua(College of Medical Engineering Technology,Xinjiang Medical University,Urumqi 830011,China;First Affiliated Hospital,Xinjiang Medical University,Urumqi 830054,China)

机构地区：[1]新疆医科大学医学工程技术学院,乌鲁木齐830011 [2]新疆医科大学第一附属医院,乌鲁木齐830011

出　　处：《科技通报》2017年第10期67-72,共6页Bulletin of Science and Technology

基　　金：2016国家自然科学基金项目(81560294);2013新疆医科大学科研创新基金项目(XJC201332)

摘　　要：目的:研究数据挖掘算法对乳腺肿瘤超声图像特征的属性选择优化,探讨适用于乳腺肿瘤良恶性分类的数据挖掘分类算法。方法:对乳腺肿瘤超声图像进行预处理,获取病灶区形状,提取病灶区图像形态、形状、纹理特征。应用数据挖掘算法进行图像特征属性选择,形成优化的乳腺肿瘤超声图像混合特征。应用分类算法评价其分类性能,筛选适用于乳腺肿瘤良恶性判定的数据挖掘分类算法。结果:利用混合特征结合随机森林算法对图像进行分类,其ROC曲线下面积AUC为0.7914,平均查准率达到了79%。结论:属性选择优化后的混合特征对乳腺肿瘤良恶性分类性能高于其他特征。在混合特征条件下,随机森林分类算法性能与Bayes网络相近,分类准确率高且性能稳定,更适于乳腺肿瘤良恶性分类评价。Objective:This paper detailed the feature optimization of ultrasound image of breast tumor usingdata mining algorithm.And investigated the data mining algorithm that was suitable for the classification of breast tumor.Methods:Firstly,the pre-processing method was applied on the ultrasound images to acquire the shapes of lesion areas.And the morphological features,shape and texture features were extracted.Secondly,the data mining algorithms were employed to select the image features and the optimal mixed features were obtained.Finally,the classification algorithms were used to evaluate the classification performance of the feature extracted above in order to find the suitable method for the classification of breast tumor.Results:Experimental results show that the AUC value,area under the ROC curve,is0.7914.And the average precision rate is79%when the random forest method and mixed features were used.Conclusions:The classification performance of the optimal mixed features is superior to other features.In comparison to Bayes method,the random forest algorithm,which has highclassification accuracy and stable performance,is more suitable to classify the breast tumor.

关 键 词：乳腺肿瘤 超声图像 图像特征 数据挖掘 属性选择 

分 类 号：R532.32[医药卫生—内科学] TP317.4[医药卫生—临床医学]