基于SITF算法的乳腺肿块分类  

Classification of Breast Masses Based on SIFT Algorithm

作　　者：王孝义 邢素霞[1] 申楠 潘子妍 WANG Xiao-yi;XING Su-xia;SHEN Nan;PAN Zi-yan(School of Computer and Information Engineering,Beijing University of Commerce and Industry,Beijing 100048,China)

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048

出　　处：《计算机仿真》2022年第1期446-450,470,共6页Computer Simulation

基　　金：国家自然科学基金(61671028);国家自然科学基金项目(61473009);北京市教委科研计划面上项目(KM201510011010);首都卫生发展科研专项(首发2018-2-5122)。

摘　　要：针对乳腺肿块特征空间复杂,分类难度大的问题,采用尺度不变特征(scale-invariant feature transform, SIFT)进行肿块特征提取,使用区域生长算法得到感兴趣区域后,使用SIFT来提取感兴趣区域的特征,再利用K-means聚类的方法得到"视觉词汇"并生成词条,将得到的词条输入SVM训练。实验结果表明,上述特征提取方法减少了描述肿块的特征维数便于分类,提高了肿块分类的准确率,为乳腺肿块计算机辅助诊断系统提供了一个新的思路。In view of the complex feature space and difficult classification of breast masses, scale invariant feature transform(SIFT) was used for mass feature extraction. After the region of interest was obtained by region growth algorithm, SIFT was used to extract the features of the region of interest, and then K-means clustering method was used to obtain “visual vocabulary”,generate entries, and input into SVM training. The experimental results show that the above feature extraction method reduces the feature dimension describing the mass, facilitates classification, improves the accuracy of mass classification, and provides a new idea for the computer-aided diagnosis system of breast mass.

关 键 词：乳腺钼靶图像 乳腺肿块 视觉词汇 支持向量机 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP391.9[自动化与计算机技术—计算机科学与技术]