摘要
目的研究纹理分析联合时间-信号强度曲线(time-intensity curve,TIC)对乳腺浸润性导管癌与纤维腺瘤鉴别诊断价值。材料与方法回顾性分析我院经病理学证实的75例女性患者(乳腺浸润性导管癌48例,乳腺纤维腺瘤27例)的MRI图像,分别绘制TIC及使用纹理分析软件中的直方图、绝对梯度、游程矩阵、共生矩阵和自回归模型共5种分析方法,对增强图像上的乳腺病灶进行纹理特征提取,共获得306个纹理特征参数;采用Fisher相关系数(Fisher coefficient,Fisher)、最小分类误判率+平均相关系数(classification error probability and average correlation coefficients,POE+ACC)及互信系数(mutual information coefficient,MI)三种统计学方法,分别筛选出区别乳腺浸润性导管癌与纤维腺瘤的10个最佳纹理特征参数。使用B11程序中的主成分分析法(principal component analysis,PCA)、线性鉴别分析法(linear discriminant analysis,LDA)和非线性鉴别分析法(nonlinear discriminant analysis,NDA)对这10个最佳纹理参数进行降维和分类,计算最佳纹理特征参数下病灶的最小误判率。统计TIC方法、纹理分析方法及两种方法联合下鉴别诊断的敏感度、特异度、准确度。结果纹理分析方法以Fisher+NDA或POE+ACC+NDA组合的最小误判率最低(4%),其筛选出用于建模的10个最佳纹理参数分别为:Fisher+NDA为GeoW1、熵S(5,-5)、相关性S(5,5)、熵S(4,-4)、熵S(5,0)、熵S(5,5)、Teta2、熵S(4,0)、Teta3、熵S(3,-3)。POE+ACC+NDA为GeoYo、Vertl_Fraction、GeoW5b、GeoW4、相关性S(5,5)、Teta1、Vertl_ShrtREmp、GeoNx、GeoAox、GeoX。TIC方法、纹理分析方法及两种方法联合下鉴别诊断的敏感度为87.5%、93.8%、97.9%;特异度为29.6%、11.1%、14.8%;准确度为66.7%、64.0%、68.0%。结论常规MRI平扫与增强的基础上,采用TIC与MRI纹理参数分析可以提高乳腺浸润性导管癌和乳腺纤维腺瘤的敏感度和准确度,其对乳腺纤维腺瘤与浸润性导管�
Objective:To study the value of texture analysis combined with time-intensity curve(TIC)in the differential diagnosis of breast invasive ductal carcinoma and fibroadenoma.Materials and Methods:The MRI images of 75 female patients(48 cases of breast invasive ductal carcinoma and 27 cases of breast fibroadenoma)confirmed by pathology were collected by retrospective method.The TIC were drawn and the five analysis methods of histogram,absolute gradient,run matrix,co-occurrence matrix and autoregressive model in texture analysis software were used to extract texture features of breast lesions on enhanced image,and obtain 306 texture feature parameters;using three statistical methods:Fisher coefficient(Fisher),classification error probability and average correlation coefficients(POE+ACC)and mutual information coefficient(MI),10 optimal texture parameters were selected to distinguish breast invasive ductal carcinoma and fibroadenoma.The principal component analysis(PCA),linear discriminant analysis(LDA)and non-linear discriminant analysis(NDA)of B11 program were used to reduce the dimension and classify the 10 optimal texture parameters,and the minimum error rate of lesions under the optimal texture parameters was calculated.The sensitivity,specificity and accuracy of TIC method,texture analysis method and the combination of the two methods were analyzed.Results:Fisher+NDA or POE+ACC+NDA combination had the lowest misjudgment rate(4%),and the 10 best texture parameters for modeling were:Fisher+NDA were GeoW1,S(5,-5)Entropy,S(5,5)Correlat,S(4,-4)Entropy,S(5,0)Entropy,S(5,5)Entropy,eta2,S(4,0)Entropy,Teta3,S(3,-3)Entropy;Poe+ACC+NDA were GeoYo,Vertl_Fraction,GeoW5b,GeoW4,S(5,5)CorrelatTeta1,Vertl_ShrtREmp,GeoNx,GeoAox,GeoX.The sensitivity of TIC,texture analysis and the combination of the two methods were 87.5%,93.8%,97.9%;the specificity were 29.6%,11.1%,14.8%;the accuracy were 66.7%,64.0%,68.0%.Conclusion:On the basis of routine MRI examination and enhancement,the sensitivity and accuracy of breast invasive ductal carci
作者
王亮
梅海清
彭红芬
张东友
韩瑞
WANG Liang;MEI Haiqing;PENG Hongfeng;ZHANG Dongyou;HAN Rui(Department of Radiology,Wuhan No.1 Hospital,Wuhan 430022,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2021年第10期53-56,共4页
Chinese Journal of Magnetic Resonance Imaging
基金
武汉市医学科研项目面上项目-重点项目(WX18B08)。
关键词
乳腺肿瘤
乳腺纤维腺瘤
乳腺浸润性导管肿瘤
鉴别诊断
磁共振成像
纹理分析
时间-信号强度曲线
breast tumor
breast fibroadenoma
breast invasive ductal carcinoma
differential diagnosis
magnetic resonance imaging
texture analysis
time-intensity curve