磁共振成像纹理特征对局部进展期直肠癌新辅助放化疗病理反应状态的预测价值

目的:探讨磁共振成像纹理特征对局部进展期直肠癌新辅助放化疗病理反应状态的预测价值。方法:回顾性分析58例局部进展期直肠癌患者的临床及影像学资料,利用MaZda软件于T2WI图像肿瘤最大层面手动勾画ROI,分别在新辅助放化疗前、后的图像上各提取9个一级、11个二级纹理特征参数,对参数进行统计学分析,比较pCR组与非pCR组在新辅助放化疗前、后纹理特征上的差别。结果:pCR组与非pCR组之间,治疗前7个二级纹理特征(pre-Correlat、InvDfMom、SumEntrp、DifEntrp、DifVarnc、Entropy、AngScMom)差异有统计学意义,治疗后3个一级纹理特征(post-Variance、Perc 90%、Perc 99%)、9个二级纹理特征(postContrast、SumOfsqs、InvDfMom、SumAverg、SumVarnc、SumEntrp、DifEntrp、DifVarnc、Entropy)差异有统计学意义。在预测pCR方面,单个纹理特征的ROC曲线结果显示,治疗前AUC值在0.632~0.835之间,治疗后AUC值在0.665~0.852之间,分别以DifVarnc、DifEntrp效能最高。采用多变量logistic回归分析预测pCR时,治疗前及治疗后的独立预测因子均是DifEntrp,治疗前(P=0.005)AUC值为0.833,治疗后(P=0.004)AUC值为0.852。结论:基于T2WI图像的纹理特征有助于预测局部进展期直肠癌新辅助放化疗的疗效,而二级纹理特征的预测效能高于一级纹理特征,nCRT治疗后为疗效预测的更佳时期。

Texture Features of Magnetic Resonance Images in Predicting the Pathological Response to Neoadjuvant Chemoradiotherapy for Rectal Cancer

Objective:To explore the performance of texture features based on T2 WI images in predicting the pathological complete response(PCR)to neoadjuvantchemoradiotherapy(nCRT)for locally advanced rectal cancer(LARC).Methods:Totally 58 patients with LARC confirmed by postoperative pathology were retrospectively analyzed.MaZda software was used to manually draw ROI on the maximum level of tumor on pre-nCRT and post-nCRT T2 WI images,and then nine first-order texture features(including Mean,Kurtosis,Skewness,Variance,Perc 1%,Perc 10%、Perc 50%、Perc 90%、Perc 99%)and eleven gray level co-occurrence matrix texture features(includingAngScMom,Contrast,Correlat,DifEntrp,DifVarnc,Entropy,InvDfMom,SumAverg,SumEntrp,SumOfSqs,SumVarnc)were extracted respectively.The texture features of each time point were statistically analyzed,and the differences of texture features between pCR and non-pCR groups were compared respectively.Results:The pre-Correlat,InvDfMom,SumEntrp,DifEntrp,DifVarnc,Entropy,AngScMom and the post-Variance,Perc 90%,Perc 99%,Contrast,SumOfsqs,InvDfMom,SumAverg,SumVarnc,SumEntrp,DifEntrp,DifVarnc,Entropydifferences between pCR and non-pCR were statistically significant.The area under the ROC curve(AUC)values for the predictors in univariate analysis ranged from 0.632 to 0.835 at the pre-nCRT stage,and the AUC values ranged from 0.665 to 0.833 at the post-nCRT stage.The results revealed that DifVarnc and DifEntrp value indicated the best efficacy for the diagnosis of pCR at each time pointrespectively.In multivariate Logistic regression analysis,pre-DifEntrp(P=0.005)and post-DifEntrp(P=0.004)were the independent predictors to pCR,with an AUC of 0.833 and 0.852 respectively.Conclusion:Texture features based on T2 WI may be potential to predict the pathological complete response of LARC receiving nCRT.The post-nCRT stage is the best time of prediction.