术前常规磁共振影像预测胶质瘤复发部位的临床研究  

Clinical study of preoperative conventional magnetic resonance imaging to predict the recurrence site of glioma

作　　者：李倩 胡晓飞 时雨 王健 LI Qian;HU Xiaofei;SHI Yu;WANG Jian(Department of Radiology,the Southwest Hospital of AMU,Chongqing 400037,China;Department of Radiology,958th Army Hospital,Chongqing 400020,China;Department of Nuclear Medicine,the Southwest Hospital of AMU,Chongqing 400037,China;Department of Pathology,the Southwest Hospital of AMU,Chongqing 400037,China)

机构地区：[1]陆军军医大学第一附属医院放射科,重庆400037 [2]陆军第九五八医院放射诊断科,重庆400020 [3]陆军军医大学第一附属医院核医学科,重庆400037 [4]陆军军医大学第一附属医院病理科,重庆400037

出　　处：《磁共振成像》2023年第8期19-26,共8页Chinese Journal of Magnetic Resonance Imaging

基　　金：国家自然科学基金(编号:92059103);四川省区域创新合作项目(编号:2023YFQ0002)。

摘　　要：目的通过术前常规MRI影像征象预测手术切除后胶质瘤复发部位(象限),从而帮助临床医师在术前规划更准确的手术切除范围。材料与方法本研究为回顾性研究,纳入了两个中心经病理确诊的123例术后复发胶质瘤患者病例,均具有完整的术前及术后复发MRI影像资料。两位放射影像科医师将术前、术后胶质瘤以其中心为原点建立平面直角坐标系,从而将肿瘤划分为四个象限,分别评估术前四个象限MRI影像征象以及术后该象限是否复发,并对两位放射影像科医师进行评分者间信度(interrater reliability,IRR)分析;选取伦勃朗视觉感受图像(Visually Accessible Rembrandt Images,VASAIR)特征集中18个MRI表现作为预测指标变量。利用二分类logistic回归作为分类器来建立预测模型,使用交叉验证的方法来验证模型的预测能力,其中训练集∶验证集=3∶1;选择有意义的变量建立列线图,并使用一致性曲线和决策曲线分析(decision curve analysis,DCA)验证。结果123例患者分别划分为四个象限后,共492个象限,将其随机分为训练集(未复发象限129个和复发象限240个)和验证集(未复发象限43个和复发象限80个),单因素分析结果发现,强化程度(P=0.03)、未强化径线(P<0.01)、深部脑白质侵犯(P=0.02)、未强化区跨中线情况(P=0.04)、室管膜侵犯(P<0.01)、T1WI/液体衰减反转恢复(fluid-attenuated inversion-recovery,FLAIR)序列(P=0.02)分布差异有统计学意义。进一步建立logistic回归模型,训练集受试者工作特征(receiver operating characteristic,ROC)曲线下面积为0.7642(P<0.05)、Kappa值为0.38,验证集数据中ROC曲线下面积为0.8493(P<0.05)、Kappa值为0.56。结论VASAIR特征集中的强化程度、未强化径线、深部脑白质侵犯、未强化区跨中线情况、室管膜侵犯、T1WI/FLAIR能够在术前预测胶质瘤复发及复发部位(象限),这对神经外科医生手术方案的制订有一定的帮助。Objective:To predict the recurrence of glioma after surgery through preoperative conventional magnetic resonance imaging signs,so as to help clinicians planning more accurate surgical resection range before surgery.Materials and Methods:This study is a retrospective study,involving 123 patients with postoperative recurrence of glioma confirmed by pathology in two centers,all of whom have complete preoperative and postoperative MRI images of recurrence.Two radiologists established a plane rectangular coordinate system with the center of the preoperative and postoperative glioma as the midpoint,thus dividing the tumor into four quadrants,respectively evaluating the MR imaging signs of the four quadrants before surgery and whether the quadrant recurred after surgery,and performing interrater reliability(IRR)analysis on the two radiologists;18 MRI manifestations of Visually Accessible Rembrandt Images(VASAIR)signs were selected as the predictive index variables.The binary logistic regression is used as a classifier to build the prediction model,and the cross-validation method is used to verify the prediction ability of the model,where the training set∶validation set=3∶1;Select meaningful variables to establish a nomogram,and use concordance index curve and decision curve analysis(DCA)to verify.Results:One hundred and twenty three patients were divided into four quadrants,a total of 492 quadrants.They were randomly divided into training set(129 non-recurrent quadrants and 240 recurrent quadrants)and validation set(43 non-recurrent quadrants and 80 recurrent quadrants).There were statistically significant differences in the enhancement quality(P=0.03),unenhanced diameter line(P<0.01),deep white matter invasion(P=0.02),unenhanced area crosses midline(P=0.04),ependymal invasion(P<0.01),the T1WI/fluid-attenuated inversion-recovery(FLAIR)(P=0.02).Further establish logistic regression model.The area under the receiver operating characteristic(ROC)curve in the training set is 0.7642(P<0.05),and the Kappa value is 0.38.Th

关 键 词：胶质瘤 磁共振成像 复发 象限 预测 

分 类 号：R445.2[医药卫生—影像医学与核医学] R730.264[医药卫生—诊断学]