基于不同扩散模型的扩散加权成像在脑胶质瘤分级和预测IDH-1突变的对比分析  

Evaluating Glioma in Terms of Grading and Predicting IDH-1 Mutation Status by Advanced Diffusion Weighted Imaging:A Comparative Study of DTI,DKI and NODDI

作　　者：黄颖倩 赵静[1] 初建平[1] 王玉亮 田一粟 丘海珊 黄子桓 HUANG Ying-qian;ZHAO Jing;CHU Jian-ping;WANG Yu-liang;TIAN Yi-su;QIU Hai-shan;HUANG Zi-huan(Department of Radiology,The First Affiliated Hospital of Sun Yat-sen University,Guangzhou 510080,China;Department of Radiology,Union Shenzhen Hospital of Huazhong University of Science and Technology,Shenzhen 518000,China;Department of Radiology,Sichuan Cancer Hospital,Chengdu 610000,China)

机构地区：[1]中山大学附属第一医院放射科,广东广州510080 [2]华中科技大学协和深圳医院放射科,广东深圳518000 [3]四川省肿瘤医院放射科,四川成都610000

出　　处：《中山大学学报（医学科学版）》2021年第1期87-94,共8页Journal of Sun Yat-Sen University:Medical Sciences

基　　金：广东省自然科学基金(2017A030313676)。

摘　　要：【目的】对比分析基于不同扩散模型的扩散加权成像(扩散张量成像:DTI,扩散峰度成像:DKI,和神经突起方向离散度与密度成像:NODDI)在脑胶质瘤术前预测脑胶质瘤级别和异柠檬酸脱氢酶-1(IDH-1)突变的诊断效能。【方法】回顾性分析中山大学附属第一医院2014年5月至2019年12月经手术病理证实的胶质瘤患者66例(WHOⅡ级28例,Ⅲ级10例,Ⅳ级28例),其中IDH-1基因表型明确的患者64例(IDH-1突变型34例,野生型30例)。采用德国西门子Magnetom Verio 3.0T MRI机器进行数据采集,扫描序列包括常规MRI平扫,增强和扩散加权成像。后处理获取DTI(平均扩散系数:MD,部分各向异性分数:FA),DKI(平均峰度:MK,轴向峰度:Ka,径向峰度:Kr)和NODDI(神经突内体积分数:icvf,神经突方向离散度:odi)的各参数图。用Image J勾画肿瘤实质相邻最大的三个层面作为ROI,取得各个参数的均值。采用独立样本t检验或Mann-Whitney秩和检验分别比较高、低级别组和不同IDH-1状态各扩散参数值的差异。进一步,对有统计学意义差异的参数进行Logistic回归分析,评价其鉴别脑胶质瘤高低级别和IDH-1基因突变状态的诊断效能,并获得受试者工作特征(ROC)曲线。【结果】基于不同的扩散加权模型,其扩散参数均可以用于区分脑胶质瘤级别(P<0.01),其中ROC分析发现MK在鉴别高低级别脑胶质瘤具有最大的诊断效能,ROC曲线下面积为0.84。进一步Logistic回归分析发现仅年龄和MK参数可以用来鉴别高低级别脑胶质瘤,诊断价值[AUC=0.88,AUC 95%CI(0.79,0.96)]优于单一的MK参数。对IDH-1基因突变状态预测,NODDI两个参数均无鉴别意义,DKI和DTI的各参数可有效鉴别(P<0.05)。其中,DKI的Ka参数具有最高的诊断价值,ROC曲线下面积为0.73,灵敏度也最高(0.83)。进一步Logistic回归发现,仅Kr可以预测IDH-1突变,回归模型ROC曲线下面积[AUC=0.72,AUC 95%CI(0.59,0.85)]。【结论】基于不同扩散模【Objective】To assess the diagnostic efficiency of different diffusion models(DTI,DKI and NODDI)in grading glioma and predicting IDH-1 mutation status,and to further build logistic regression prediction models.【Methods】Totally 66 patients(22 females;mean age:47.8)with pathologically proved gliomas were retrospectively included.All cases underwent bipolar spin echo diffusion examination.Parameters of DKI(MK;Ka;Kr),DTI(MD and FA)and NODDI(intracellular volume fraction:icvf,orientation dispersion index:odi)were derived.ROIs were manually drawn and corresponding average values were calculated.Logistic regression was performed to build a predictive model.ROC curve was obtained,and Hosmer-lemeshow test was carried out to test the goodness of fit.【Results】DKI,DTI and NODDI parameters were significantly different between HGGs and LGGs(P<0.01).And among all diffusion parameters,a further logistic regression model for grading glioma only included age and MK,which showed the highest diagnostic value[AUC=0.88,AUC 95%CI(0.79,0.96)].Hosmer-lemeshow Test present excellent of goodness of fit.With IDH-1 mutation status,NODDI showed no significant value for distinction,whereas DKI and DTI can significantly differentiate IDH-1 mutated and non-mutated glioma(P<0.05).Further logistic regression only selected Kr(P<0.01)in the model,which demonstrated the highest diagnostic value[AUC=0.72,AUC 95%CI(0.59,0.85)].【Conclusions】DKI is superior to DTI and NODDI in grading gliomas and identifying IDH-1 mutation status.The model of MK value and age variables present the best discriminatory capacity for grading glioma and Kr value may serve as a potential predictive index for identify IDH-1 mutation.

关 键 词：脑 胶质瘤 异柠檬酸脱氢酶-1 磁共振成像 分子亚型 

分 类 号：R739.41[医药卫生—肿瘤]