治疗前IVIM-DWI参数图纹理特征对直肠癌新辅助放化疗病理反应的预测价值

Texture features derived from intravoxel incoherent motion diffusion-weighted imaging for predicting the pathological response to chemoradiotherapy in rectal cancer

目的探讨基于体素内不相干运动扩散加权成像(intravoxel incoherent motion diffusion-weighted imaging,IVIM-DWI)参数图的纹理特征对直肠癌原发灶新辅助放化疗病理反应的预测价值。材料与方法搜集38例直肠癌患者的影像及病理资料,比较其中原发灶病理完全反应(pathological complete response,pCR)组与非pCR组在治疗前IVIM-DWI参数(ADC、D、D*及f)图纹理特征上的差别。结果与非pCR组相比,pCR组的AngScMom_D、AngScMom_(D*)、AngScMom_f、DifVarnc_(ADC)、DifVarnc_D、Contrast_(ADC)及Contrast_D*值更低,而Perc10%_(ADC)、Perc10%_D、Perc99%_(D*)、Correlat_(D*)、Correlat_f、DifEntrp_(ADC)、InvDfMom_(ADC)、SumAvergD、SumVarnc_(D*)及SumOfSqs_(D*)值更高。在预测pCR方面,单个纹理特征的ROC曲线下面积(area under curve,AUC)值为0.662~0.829。采用多变量(一级+二级纹理特征)Logistic回归模型预测pCR时,DifVarnc_D(P=0.003)和SumVarnc_(D*)(P=0.002)为独立预测因子,AUC值为0.929。结论基于IVIM-DWI参数图的纹理特征可能有助于预测直肠癌原发灶新辅助放化疗病理反应。

Objective:To investigate the performance of texture features based on intravoxel incoherent motion diffusion-weighted imaging(IVIM-DWI)on identifying pathological complete response(pCR)to neoadjuvant chemoradiotherapy(nCRT)in locally advanced rectal cancer(LARC).Materials and Methods:Pretreatment IVIM-DWI was performed on 38 LARC patients receiving nCRT.Nine first-order texture features(TFs)and eleven gray level co-occurrence matrix(GLCM)TFs were derived from four IVIM-DWI parameter maps(ADC,D,D*and f)respectively.The first-order TFs included Mean,Kurtosis,Skewness,Variance,Perc01%,Perc10%,Perc50%,Perc90%and Perc99%,and the GLCM features included Angular Second Moment(AngScMom),Contrast,Correlat,Difference Entropy(DifEntrp),Difference Variance(DifVarnc),Entropy,Inverse Difference Moment(InvDfMom),Sum Average(SumAverg),Sum Entropy(SumEntrp),Sum of Squares(SumOfSqs)and Sum Variance(SumVarnc).The values of first-order and GLCM TFs were compared between the pCR(n=8)and non-pathological responder(non-pCR,n=30)groups,which was classified according to tumor regression grade system.Receiver operating characteristic(ROC)curve in univariate and multivariate Logistic regression analysis was generated to determine the efficiency for identifying pCR.Results:The pCR group had lower AngScMomD,AngScMomD*,AngScMomf,DifVarncADC,DifVarncD,ContrastADC and ContrastD*values.Higher Perc10%ADC,Perc10%D,Perc99%D*,CorrelatD*,Correlatf,DifEntrpADC,InvDfMomADC,SumAvergD,SumVarncD*and SumOfSqsD*values were observed in the pCR group.The area under the ROC curve(AUC)values for the predictors in univariate analysis ranged from 0.662 to 0.829,with sensitivities from 33.33%to 100.00%and specificities from 37.50%to 100.00%.In multivariate Logistic regression analysis based on the first-order TFs,Perc10%ADC(P=0.032)and Perc10%D(P=0.028)were the independent predictors to pCR,with an AUC value of 0.754(95%confidence interval,0.588-0.879),a sensitivity of 50%and a specificity of 100.00%.DifVarncD(P=0.003)and SumVarncD*(P=0.002)were the independent predictors to pCR in the multivariate models that were based on either the GLCM TFs or the combination of the first-order and GLCM TFs,with an AUC of 0.929(95%confidence interval,0.797-0.987),a sensitivity of 83.33%and a specificity of 100.00%.Conclusions:GLCM analysis based on IVIM-DWI may be a potential approach to identify the pathological response of LARC before starting chemoradiotherapy.