MRI纹理分析在单囊状胰腺囊腺瘤鉴别诊断中的价值  

Value of MRI radiomics for the differential diagnosis of unicystic pancreatic cystadenoma

作　　者：王希恒 薛华丹[1] 成思航 李娟[1] 孙照勇[1] 金征宇[1] WANG Xi-heng;XUE Hua-dan;CHENG Si-hang(Department of Radiology,Peking Union Medical College Hospital,Peking Union Medical College,Chinese Academy of Medical Sciences,Beijing 100730,China)

机构地区：[1]中国医学科学院/北京协和医学院/北京协和医院放射科,北京100730

出　　处：《放射学实践》2021年第6期756-761,共6页Radiologic Practice

基　　金：首都卫生发展科研专项基金资助项目(2018-2-4014);国家重点研发计划资助(2020YFC2002702)。

摘　　要：目的:比较胰腺寡囊型浆液性囊腺瘤(MaSCA)与粘液性囊腺瘤(MCN)磁共振成像(MRI)影像学及纹理特征差异,构建两者鉴别诊断模型。方法:回顾性搜集32例MaSCA与36例MCN患者,分别基于MRI压脂T_1加权图像(FS-T_1WI)和压脂T_2加权图像(FS-T_2WI)进行纹理分析并比较各纹理参数之间差异。采用Logistic回归分析分别对具有差异影像特征构建影像模型,选取曲线下面积(area under the curve, AUC)最大参数构建纹理分析模型,两组特征共同构建组合模型,利用ROC曲线(receiver-operating characteristic curve)评估模型诊断效能。结果:影像模型AUC 0.849,敏感度及特异度分别为86.1%、68.7%;纹理分析模型AUC 0.887,敏感度及特异度均较高,分别约80.6%、84.4%。组合模型AUC最高,为0.958,敏感度及特异度分别为88.9%、90.6%。结论:综合影像特征和纹理分析特征组合模型,有助于术前鉴别MaSCA和MCN且具有很高诊断性能。Objective:To compare the differences of magnetic resonance imaging(MRI)and texture characteristics between pancreatic macrocystic serous cystadenoma(MaSCA)and mucinous cystic neoplasm(MCN),and furthermore,to develop the preoperative differentiation models.Methods:Six-eight patients(MaSCA=32,MCN=36)with preoperative MRI scans using fat suppression T 1 weighted imaging(FS-T 1WI)and fat suppression T 2 weighted imaging(FS-T 2WI)were retrospectively analyzed in this study.Texture characteristics analysis based on FS-T 1WI and FS-T 2WI images were compared between MaSCA and MCN.Logistic regression analysis was used to construct image models with different image features,and the texture characteristics analysis model was constructed by selecting the maximum area under the curve(AUC).The combined model was constructed by two groups of features,and the diagnostic efficiency of the model was evaluated by ROC curve.Results:The radiological model of yielded an AUC of 0.849,and sensitivity and specificity were 81.6%and 68.7%respectively.The radiomics model yielded an AUC of 0.887 with relatively higher sensitivity and specificity of 80.6%and 84.4%respectively.The combined model yielded a highest AUC of 0.958,and sensitivity and specificity were 88.9%and 90.6%respectively.Conclusion:The combined model based on image features and texture analysis features is helpful to distinguish Masca and MCN before operation,and shows high diagnostic performance.

关 键 词：胰腺 囊腺瘤 浆液 囊腺瘤 黏液 磁共振成像 

分 类 号：R322.491[医药卫生—人体解剖和组织胚胎学] R730.269[医药卫生—基础医学]