构建磁共振影像组学模型鉴别诊断乳腺浸润性导管癌及乳腺硬化性腺病价值  

Value of the model of magnetic resonance imaging omics for differential diagnosing breast infiltrating ductal carcinoma and breast sclerosing adenosis

作　　者：赵新凯 王茂林[2] 郑小敏 ZHAO XinKai;WANG Maolin;ZHENG Xiaomin(Graduate school,Zhejiang Chinese Medical University,Hangzhou,Zhejiang Province,310053;Wuxi Maternity and Child Health Care Hospital,Wuxi)

机构地区：[1]浙江中医药大学研究生院,杭州310053 [2]江苏省无锡市妇幼保健院

出　　处：《中国计划生育学杂志》2023年第10期2348-2353,2522,共7页Chinese Journal of Family Planning

摘　　要：目的:探索磁共振MRI影像组学在乳腺浸润性导管癌(IDC)与乳腺硬化性腺病(SA)鉴别诊断中的价值。方法:收集2020年7月-2022年11月经手术病理证实为IDC42例和SA18例临床资料,分为训练组42例和测试组18例,磁共振仪获取患者乳腺MRI图像,分析MRI影像学特点及其诊断价值。患者均采用乳腺专用7通道线圈行磁共振双侧乳腺平扫及动态增强扫描。使用3D Slicer软件对T2WI、DWI、ADC及T1WI增强第3期及第7期序列图像的感兴趣区进行手动分割,并提取纹理特征、计算影像组学积分和创建影像组学模型,绘制各模型的受试者工作曲线(ROC)和绘制决策分析曲线评估模型的的预测效能和临床应用价值。结果:通过对T2WI、DWI、ADC以及T1WI增强第3期和第7期5个序列两次降维筛选后最终获得38个最优纹理特征(一阶统计特征4个、形态特征9个、GLCM特征13个、GLSZM特征3个、NGTDM特征2个、GLDM特征4个、GLRLM特征3个),采用logistic回归构建5个诊断模型,计算获得了预测IDC和SA病理分型的影像组学积分。ROC分析训练组中5个模型曲线下面积(AUC)分别为0.976、0.969、0.964、0.997、0.997,灵敏度分别为0.937、0.900、1.000、0.967、0.967,特异度分别为0.967、1.000、0.917、1.000、1.000;测试组中5个模型AUC分别为0.986、0.944、0.896、0.944、0.972,灵敏度分别为0.917、1.000、1.000、0.833、0.917,特异度分别为1.000、0.833、0.857、1.000、1.000。训练组和测试组校准曲线显示该模型均具有良好的校准度,决策曲线提示所构建的影像组学模型对IDC和SA有良好的鉴别诊断效能。结论:基于5个序列(T2WI、DWI、ADC、T1WI增强第3期及第7期)所构建的影像组学模型对IDC和SA有良好的鉴别诊断效能。Objective:To explore the value of the model of magnetic resonance imaging omics for differential diagnosing breast infiltrating ductal carcinoma(IDC) and breast sclerosing adenosis(SA) of patients.Methods:From July 2020 to November 2022,the clinical data of 42 patients with ID(in group A) and 18 patients with SA(in group B) were collected.The MRI of the patients in the two groups was obtained by magnetic resonance instrument,and the features and diagnostic value of MRI of the patients were analyzed.All the patients underwent the bilateral mammography performed by a special 7-channel coil MRI and the dynamic enhanced scans.The regions of interest of T2WI,DWI,ADC,and T1WI enhanced phase 3 and phase 7 sequence images of the patients were manually segmented by 3D Slicer software,and the texture features was extracted,the imaging omics integrals was calculated,and the imaging omics model was established.Receiver operator characteristic(ROC) curve and decision analysis curve were drawn to evaluate the predictive efficacy and clinical application values of this model.Results:The results showed that 38 optimal texture features(4 first-order statistical features,9 morphological features,13 GLCM features,3 GLSZM features,2 NGTDM features,4GLDM features,and 3GLRLM features)were ultimately obtained after two dimensionality reduction screenings of 5sequences enhanced with T2WI,DWI,ADC,and T1WI in phases 3and 7.5diagnostic models were constructed by Logistic regression,and the image omics scores of predicting the pathologic classification of IDC and SA were calculated.In ROC analysis,the area under the curve(AUC)of the 5models for predicting the pathologic classification of IDC and SA of the patients in group A were 0.976,0.969,0.964,0.997,and 0.997,respectively,the sensitivity of which were 0.937,0.900,1.000,0.967and 0.967,respectively,and the specificity of which were was0.967,1.000,0.917,1.000,and 1.000,respectively.The AUC of the 5models for predicting the pathologic classification of IDC and SA of the patients in group B

关 键 词：乳腺硬化性腺病 乳腺浸润性导管癌 乳腺X线摄影 超声 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]