乳腺肿瘤超声图像的特征分析  

Feature analysis of ultrasound breast tumor images

作　　者：林其忠[1] 余建国[1] 赵暖[1] 王威琪[1] 王怡[2] 陈亚青[3] 

机构地区：[1]复旦大学电子工程系,上海200433 [2]复旦大学附属华山医院,上海200040 [3]上海市第六人民医院,上海200233

出　　处：《仪器仪表学报》2006年第z1期744-746,752,共4页Chinese Journal of Scientific Instrument

基　　金：上海市科学技术委员会科研计划项目资助(054119612)

摘　　要：基于乳腺肿瘤良恶性在超声图像的不同特征,利用计算机自动识别,作为医生的辅助诊断。方法的步骤为:本文先在常用超声仪上获得乳腺肿瘤超声图像,接着从图像中自动提取肿瘤边缘,然后自动提取不依赖于超声仪系统的特征参数,用特征选择器选择出最优特征矢量,最后经分类器判别乳腺肿瘤的良恶性。实验基于200例病例随机划分为训练集和测试集各半进行测试,获得结果Accuracy为0.960,Sensitivity为0.982,Specificity为0.935,PPV和NPV分别为0.946和0.977,结果表明本文方法泛化能力强,可以作为识别乳腺肿瘤良恶性的一种辅助手段。To develop a computer-aided diagnosis with multiple feature to differentiate benign from malignant breast tumor.Ultrasound Breast Tumor Image was firstly obtained using a general ultrasonic scanner.From this image,Tumor boundary was extracted automatically,Then optimal feature vector was selected from features with nearly independent on setting extracted from Ultrasound Breast Tumor Image using Sequential Forward Selection Algorithm.Finally SVM classifier was used to differentiate benign from malignant breast tumor.Experiments on 200 ultrasonic images,randomly divided into training set 100 and prediction set 100,show that Accuracy was 0.960,Sensitivity was 0.982,Specificity was 0.935,PPV was 0.946 and NPV was 0.935.The proposed algorithm has better generalization and could effectively differentiate benign and malignant lesions.

关 键 词：超声图像 特征提取 特征选择 分类器 

分 类 号：TH7-55[机械工程—仪器科学与技术]