MRI常规特征及纹理特征对乳腺乳头状病变的鉴别诊断价值  

Value of MRI conventional features and texture features in the differential diagnosis of papillary lesions of the breast

作　　者：方舒 章伟 朱俊[1] 柳婷艳 丁茹 王桂钱 FANG Shu;ZHANG Wei;ZHU Jun;LIU Tingyan;DING Ru;WANG Guiqian(Department of Radiology,Changzhou Hospital of Traditional Chinese Medicine,Nanjing University of Traditional Chinese Medicine,Changzhou,Jiangsu Province 213000,China)

机构地区：[1]南京中医药大学附属常州市中医医院放射科,江苏常州213000

出　　处：《实用放射学杂志》2022年第12期1962-1966,共5页Journal of Practical Radiology

基　　金：常州市卫生健康青苗人才培养工程项目(CZQM2020082);常州市卫生健康委员会科技项目(WZ202207)。

摘　　要：目的探讨MRI常规特征及纹理特征对乳腺良恶性乳头状病变的鉴别诊断价值。方法纳入经手术病理证实的116例患者共124个乳头状病变,观察记录患者的年龄、病灶部位、强化方式、是否伴有导管扩张、是否伴有囊性灶、病灶形态、病灶时间信号强度曲线(TIC)类型、病灶大小及表观扩散系数(ADC)值。采用3D Slicer软件对第2期动态增强图像勾画感兴趣区(ROI),并用其集成插件SliceRadiomics提取纹理特征参数。应用独立样本t检验及卡方检验比较2组间MRI常规特征及纹理特征的差异,对筛选出的有统计学意义的特征进行多因素Logistic回归分析,并建立回归模型。采用受试者工作特征(ROC)曲线评估各模型的诊断效能。结果良性病灶69个,恶性病灶55个。在临床及MRI常规特征中,年龄和病灶形态是预测乳腺乳头状恶性病变的独立危险因素。在提取的93个纹理特征中,筛选出4个特征参数:Size Zone Non Uniformity Normalized、Correlation、Short Run Emphasis、Small Dependence Low Gray Level Emphasis。所建立的临床及MRI常规特征模型、纹理特征模型及联合模型的曲线下面积(AUC)分别为0.744、0.708、0.862,联合模型的敏感度89.1%,特异度81.2%,较前两者高。结论在常规MRI的基础上,采用纹理特征与MRI常规特征相结合的方法可以提高对乳腺乳头状良恶性病变的诊断效能,有助于更好地指导临床医师对乳腺乳头状病变做出决策。Objective To explore the value of MRI conventional features and texture features in the differential diagnosis of benign and malignant papillary lesions of the breast.Methods A total of 124 papillary lesions in 116 patients confirmed by surgery and pathology were included.Patients’age,lesion site,enhancement pattern,ductal dilatation,whether there was cystic foci,lesion shape,lesion time-signal intensity curve(TIC)type,lesion size and apparent diffusion coefficient(ADC)value were observed and recorded.The region of interest(ROI)were drawn by using 3D Slicer software and its integrated plug-in SliceRadiomics was used for texture features extraction in the second phase of the dynamic enhanced image.Independent sample t-test and Chi-square test were used to compare the differences in MRI conventional features and texture features between the two groups.Multivariate Logistic regression analysis was performed on the statistically significant features,and regression models were further established.The receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficiency of each model.Results There were 69 benign lesions and 55 malignant lesions.Age and lesion shape were the independent risk factors for predicting malignant papillary lesions of the breast in the clinical and conventional features.Of all 93 texture features,4 features parameters were extracted,including Size Zone Non Uniformity Normalized,Correlation,Short Run Emphasis,and Small Dependence Low Gray Level Emphasis.The area under the curve(AUC)of the clinical and MRI conventional features model,texture features model and combined model were 0.744,0.708,0.862,respectively.The sensitivity of the combined model was 89.1%,and the specificity was 81.2%,which were higher than the former two.Conclusion The combination of texture features and MRI conventional features can improve the diagnostic efficiency of benign and malignant papillary lesions of the breast,and better provide options for clinical therapy.

关 键 词：乳腺 乳头状病变 纹理分析 常规特征 磁共振成像 

分 类 号：R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学]