基线磁共振T2WI纹理预测晚期直肠癌转化治疗原发灶疗效的应用研究

Application study of MRI T2WI texture baseline predicting the efficacy of advanced rectal cancer transformation therapy for primary tumors

目的探讨基线磁共振成像(magnetic resonance imaging,MRI)T2WI图像纹理分析在晚期直肠癌转化治疗原发灶疗效的预测价值。材料与方法回顾性分析临床及病理证实为晚期直肠癌的患者66例,基线行盆腔MRI平扫、增强及扩散加权成像(diffusion weighted imaging,DWI)检查。根据平扫及增强图像明确肿瘤的部位及范围,运用Mazda软件提取T2WI图像感兴趣区(region of interest,ROI)纹理,分别运用线性判别分析(linear discriminant analysis,LDA)、非线性判别分析(nonlinear discriminant analysis,NDA)和主要成分分析(principal component analysis,PCA)3种提取方法进行判别分类,筛选出最优方法进行纹理提取。结合术后病理,比较晚期直肠癌患者原发灶疗效敏感组与不敏感组基线形态学特征,比较两组T2WI序列图像纹理特征,构建疗效预测模型。结果66例晚期直肠癌患者原发灶术后病理肿瘤退缩分级(pathological tumor regression grade,pTRG)显示:pTRG 0级9例,pTRG 1级8例,pTRG 2级35例,pTRG 3级14例,其中敏感组(pTRG 0~2级)52例和不敏感组(pTRG 3级)14例。两组患者原发灶累及肠段、与腹膜反折关系、纵向累及长度、占肠腔环周比例、斜轴位最大厚度、肿瘤下缘距肛缘的距离差异均无统计学意义(P均>0.05);Fisher纹理特征提取法下的NDA分类方法误判率最低,故运用该方法提取图像纹理。晚期直肠癌转化治疗原发灶不同疗效组别各纹理特征单因素分析显示:第一百分位数(Percentile,Perc 1%)、S(2,0)DifEntrp、S(3,0)InvDfMom、S(3,-3)SumAverg、S(4,0)InvDfMom、S(4,-4)SumAverg、S(5,0)InvDfMom、S(5,-5)SumAverg、S(2,2)SumVarnc各指标差异均有统计学意义(P均<0.05),S(2,2)SumVarnc、S(3,0)DifEntrp差异无统计学意义(P=0.05、0.052);将单因素分析差异有统计学意义的指标纳入Logistic模型进行多因素分析显示:Perc 1%、S(5,0)InvDfMom为晚期直肠癌原发灶转化治疗不敏感的独立预测因子,运用上述因子构建晚期直肠癌原发灶转化治疗不敏感预测模型曲线下面积(area under the curve,AUC)为0.812,敏感度为92.90%,特异度为60.80%。结论基于MRI Fisher提取法所提取的T2WI图像纹理特征有助于预测晚期直肠癌原发灶转化治疗疗效,为患者个体化治疗方案的制定提供有价值的参考信息。

Objective:To explore the predictive value of baseline magnetic resonance imaging(MRI)T2WI image texture analysis in the treatment of advanced rectal cancer for the primary tumor.Materials and Methods:Retrospective analysis of 66 patients with advanced rectal cancer confirmed clinically and pathologically.All patients underwent pelvic MRI scan,enhancement and diffusion weighted imaging(DWI)examinations before operation.According to the plain scan and enhanced images,the location and range of the tumor were identified,and the Mazda software was used to extract the region of interest(ROI)texture in the T2WI image,and linear discriminant analysis(LDA)and nonlinear discriminant analysis(linear discriminant analysis)were used respectively.Discriminant analysis(NDA)and principal component analysis(PCA)are three extraction methods for discriminative classification,and the best method is selected for texture extraction.Combined with postoperative pathology,the baseline morphological characteristics of the primary focus of patients with advanced rectal cancer were compared between the sensitive group and the insensitive group,and the texture characteristics of the T2WI sequence images of the two groups were compared to construct a curative effect prediction model.Results:The pathological tumor regression grade(pTRG)of 66 patients with advanced rectal cancer showed that 9 cases were pTRG 0,8 cases were pTRG 1,35 cases were pTRG 2,and 14 cases were pTRG 3.Among them,52 cases were in the sensitive group(pTRG 0~2)and 14 cases were in the insensitive group(pTRG 3).There was no significant difference between the two groups of patients between the primary tumor involving the intestinal segment,the relationship with the peritoneum reflexion,the length of the longitudinal involvement,the proportion of the circumference of the intestinal cavity,the maximum thickness of the oblique axis,and the distance between the lower edge of the tumor and the anal edge(all P>0.05);the NDA classification method under the Fisher texture feature extraction method has the lowest misjudgment rate,so this method is used to extract the image texture.The univariate analysis of texture characteristics in different treatment groups of the primary tumor of advanced rectal cancer showed:the first percentile(Perc 1%),S(2,0)DifEntrp,S(3,0)InvDfMom,S(3,-3)SumAverg,S(4,0)InvDfMom,S(4,-4)SumAverg,S(5,0)InvDfMom,S(5,-5)SumAverg,S(2,2)SumVarnc,all indicators were statistically different(P<0.05),S(2,2)SumVarnc,S(3,0)DifEntrp were not statistically different(P=0.05,0.052);the indicators with differences in univariate analysis were included in Logistic multivariate analysis of the model showed that Perc 1%and S(5,0)InvDfMom were independent predictors of insensitivity to transformation treatment of primary tumors of advanced rectal cancer,and the above factors were used to construct the prediction of insensitivity of primary tumors of advanced rectal cancer to transformation therapy.The area under the curve(AUC)of the model is 0.812,the sensitivity is 92.90%,and the specificity was 60.80%.Conclusions:T2WI image texture features extracted based on MRIFisher extraction method can help predict the efficacy of primary tumor transformation therapy for advanced rectal cancer,and provide valuable reference information for the formulation of individualized treatment plans for patients.