基于全肿瘤ADC图纹理特征预测术前WHOⅡ~Ⅲ级别胶质瘤IDH1基因表达类型效能的初步探讨

Prediction efficacy of the whole tumor volume ADC-derived texture features for IDH1 mutation status of preoperative WHOⅡ~Ⅲgrade glioma

目的:探讨基于全肿瘤ADC图的纹理特征预测术前WHOⅡ~Ⅲ级别原发脑胶质瘤IDH1基因类型的效能。方法:对99例WHOⅡ~Ⅲ级别原发胶质瘤患者行T1WI、T2WI、液体反转恢复序列(FLAIR)、DWI及增强T1WI扫描。基于术前ADC图,提取包括10个一阶直方图特征和9个灰度共生矩阵(GLCM)特征在内的19个全肿瘤体积纹理特征,结合常规MRI形态特征,比较IDH1基因突变组与野生组间各影像特征的差异,并采用Logistic回归模型分析差异有统计学意义的变量,筛选出IDH1突变类型的独立预测因子,绘制ROC曲线,比较其与ADC平均值、最小值的诊断效能。结果:不同IDH1基因表达类型组间,单因素分析结果显示单/多发、增强程度及8个纹理特征(最小值、10%百分位数、25%百分位数、方差、峰度、GLCM熵、GLCM平均和、GLCM方差)差异具有统计学意义(P<0.05)。Logistic回归分析结果显示增强程度、最小值、GLCM熵为IDH1基因突变类型的独立预测因子。联合增强程度、最小值、GLCM熵诊断的敏感度为84.0%,特异度为77.6%,AUC达0.845,较ADC平均值(AUC=0.585)及最小值(AUC=0.730)具有显著差异(P<0.05)。结论:常规MRI形态特征联合基于全肿瘤ADC图纹理特征的诊断模式能从定量角度提高术前脑胶质瘤IDH1基因突变状态的预测效能。

Objective:To explore the prediction efficacy of the whole tumor volume ADC-derived texture features for IDH1 mutation status of preoperative WHOⅡ~Ⅲgrade glioma.Methods:Ninety-nine patients with WHOⅡ~Ⅲglioma performed T1WI,T2WI,T2 fluid-attenuated inversion recovery(FLAIR)sequence,diffusion-weighted imaging(DWI),and contrastenhanced T1WI were retrospectively enrolled.Nineteen texture features were extracted from ADC maps of the whole tumor,including first-order histogram(10 features)and gray-level co-occurrence matrix(GLCM,9 features).MRI morphological and texture features were compared between IDH1 mutant and wild-type gliomas using the Pearson Chi-Square test,the non-parametric Wilcoxon rank-sum test or the unpaired Student’s t-test.Variates with statistical significance at univariate analysis were then screened by Logistic regression analysis.We used receiver characteristic(ROC)analysis to evaluate diagnostic performance of independent predictors,mean and minimum ADC values.Results:Significant differences were detected at univariate analysis between IDH1 mutant and wild-type gliomas for number,enhancement and 8 texture features(Histogram min,Histogram perc10,Histogram perc25,Histogram variance,Histogram kurtosis,GLCM entropy,GLCM sumaverage,GLCM variance).Logistic analysis showed that enhancement,Histogram min and GLCM entropy were independent predictors for IDH1 mutation status.Among all factors,combining enhancement with Histogram min and GLCM entropy showed a higher efficacy(AUC=0.845),with a sensitivity and specificity of 84.0%and 77.6%,respectively.Statistically significant differences were found for the AUC values between the combination model,mean ADC value(AUC=0.585)and minimum ADC values(AUC=0.730)(P<0.05).Conclusion:The diagnostic model that combining texture features derived from whole tumor volume ADC mapping with conventional MRI morphological features,can quantitatively improve the predictive performance of IDH1 gene expression type in preoperative WHOⅡ~Ⅲgrade glioma.