乳腺肿瘤超声图像纹理特征分析  

Textural Feature Analysis for Ultrasound Breast Tumor Images

作　　者：陈秋霞[1] 向军[1] 刘奇[2] 刘剑[1] 

机构地区：[1]重庆医科大学附属第一医院设备处,重庆400016 [2]四川大学电气信息学院医学信息工程系,成都610065

出　　处：《激光杂志》2014年第3期65-66,69,共3页Laser Journal

基　　金：西华大学四川省信号与信息处理重点实验室基金项目(SGXZD0101-10-1)

摘　　要：提取乳腺肿瘤超声图像的肿瘤区域,计算乳腺肿瘤图像的纹理特征参数,研究纹理特征与肿瘤良恶性的关系。基于综合空间灰度共生矩阵计算11个乳腺肿瘤超声图像的纹理特征参数,然后分别利用模糊C均值和K-medoid聚类算法对乳腺肿瘤进行良恶性判别,同时,通过重复实验找到判别肿瘤良恶性的最佳特征参数组合。实验结果表明相关性、和方差、相关信息度量1和相关信息度量2四个特征参数组合的判别结果最好,达到了72.64%。因此,纹理特征在一定程度上能够反映良恶性乳腺肿瘤的区别,其对于鉴别乳腺肿瘤的良恶性是有效的。To extract the tumor area of the breast tumors in ultrasonography, upon which calculate the textural features and study the relationship between textural features and pathological nature of the tumors. Eleven textural features are calculated based on the in-tegrated spatial gray level co-occurrence matrix, and the Fuzzy C-means and K-medoid algorithms are applied to differentiate the be-nign and the malignant breast tumors respectively. Meanwhile, the best combination of textural features is determined through repeat-ed experiment. It shows that the combination of feature correlation, sum variance, information measure of correlation 1 and informa-tion measure of correlation 2 can bring the best differentiation result with accuracy of 72.64%. Therefore the textural features can re-flect the difference of benign and malignant breast tumors to some extent, and which is effective in differentiating the benign and ma-lignant breast tumors.

关 键 词：乳腺肿瘤 纹理特征 聚类算法 

分 类 号：TN244.8[电子电信—物理电子学]