超声散射子特性成像评估乳腺肿瘤良恶性研究  

Classification of benign and malignant breast tumors using ultrasound scatterer property imaging

作　　者：欧阳亚丽 吴水才[1] 崔博翔 周著黄[1] OUYANG Ya-li;WU Shui-cai;TSUI Po-hsiang;ZHOU Zhu-huang(College of Life Science and Bioengineering,Beijing University of Technology,Intelligent Physiological Measurement and Clinical Translation,Beijing International Base for Scientific and Technological Cooperation,Beijing 100124,China;College of Medicine,Chang Gung University,Taoyuan 33302,Taiwan Province,China)

机构地区：[1]北京工业大学生命科学与生物工程学院,智能化生理测量与临床转化北京市国际科研合作基地,北京100124 [2]台湾长庚大学医学院,中国台湾桃园33302

出　　处：《医疗卫生装备》2019年第9期1-7,共7页Chinese Medical Equipment Journal

基　　金：北京市自然科学基金项目（4184081）;国家自然科学基金项目（61871005,61801312);19内涵发展定额-筑基工程-博士科研启动金

摘　　要：目的:提出一种基于超声散射子特性成像评估乳腺肿瘤良恶性的新方法,并验证其分类效果。方法:选取78位女性乳腺肿瘤患者的超声射频数据分别进行H-Scan超声成像,其中,良性病灶48例、恶性病灶52例。H-Scan超声成像利用高斯加权Hermite多项式(GHn)进行匹配滤波。利用GH2和GH8滤波核对超声射频数据进行并行卷积滤波,将GH2滤波成像结果和GH8滤波成像结果分别分配给红色通道R和蓝色通道B,原始射频数据成像结果分配给绿色通道G,根据R、G、B通道,实现H-Scan超声成像。采用SPSS 19.0软件进行数据统计分析。结果:良性乳腺肿瘤病灶区域显示出更多的红色分量,而恶性乳腺肿瘤病灶区域与良性乳腺肿瘤病灶区域相比,蓝色分量更为显著。良恶性乳腺肿瘤R、G、B通道统计分析发现,每个H-Scan成像结果所有像素相对应的R、B通道差值求均值的统计结果(t=3.275,P<0.05)、R通道和B通道的均值之差与G通道的比值统计结果(t=3.275,P<0.05)以及R通道平均值与B通道平均值之间的比值统计结果(t=3.275,P<0.05)均具有显著性差异。采用ROC曲线分别对R、G、B通道的均值进行统计分析发现,R、G、B通道曲线下面积分别为0.838、0.811、0.765,P值均小于0.001。结论:H-Scan超声成像能够有效区分良恶性乳腺肿瘤病灶,其中,良性乳腺肿瘤病灶区域多为直径较大的散射子,而恶性乳腺肿瘤病灶多为直径较小的散射子。Objective To propose a new classification method of benign and malignant breast tumors based on ultrasound scatterer property imaging, and to analyze the performance of the proposed method. Methods The radiofrequency data of 78 female breast tumor patients were selected for H-Scan ultrasound imaging, including 48 benign cases and 52 malignant cases. H-Scan ultrasound imaging could image the relative size of scatterers by matched filtering using Gauss-weighted Hermite polynomial(GHn). The GH2 and GH8 filters were used to perform parallel convolution filtering on the ultrasonic radiofrequency data. The GH2 filter imaging results and GH8 filter imaging results were assigned to the red channel(R) and blue channel(B), respectively, and the original radiofrequency data imaging results were assigned to the green channel(G).H-Scan ultrasound imaging was achieved based on R, G and B channels. SPSS 19.0 software was used for data analysis and statistics, and P <0.05 indicated statistical significance. Results Benign breast tumors showed more red components, while malignant breast tumors showed more blue components. For the statistical analysis of R, G and B channels, the results of average value calculated by the minus of R, B channel corresponding to each pixel for all images(t=3.275, P<0.05), the ratio result of the mean of R channel minus the mean of B channel with G channel(t=3.275, P<0.05), and the ratio of R channel to B channel(t =3.275, P <0.05), all had significant differences. Mean analysis of each channel by the receiver operating characteristic(ROC) curve showed that the area under the ROC curve(AUC) of R, G and B channels was 0.838, 0.811, and0.765, respectively, and the P values were all less than 0.001. Conclusion H-Scan ultrasound imaging can effectively distinguish benign and malignant breast lesions. Benign breast cancer lesions with larger scatterers are more red in hue, while malignant breast cancer lesions with smaller scatterers are more blue in hue.

关 键 词：乳腺肿瘤 超声成像 H-Scan超声成像 HERMITE多项式 散射子大小 超声背散射 超声组织定征 

分 类 号：R318[医药卫生—生物医学工程] R445.1[医药卫生—基础医学]