直肠癌DCE-MRI纹理分析预测Ki67表达状态的价值

DCE-MRI texture analysis of rectal cancer to predict the expression of Ki67 state value

目的:探讨动态对比增强磁共振成像(Dynamic contrast-enhanced magnetic resonance imaging,DCE-MRI)纹理分析术前预测直肠癌患者Ki67表达状态的价值。方法:回顾性收集行DCE-MRI扫描且有Ki67免疫组化检测结果的原发直肠癌患者91例(Ki67高表达组56例、低表达组35例)。由GenIQ后处理软件获取K^(trans)、K_(ep)、V_(e)后处理图像导入ITK-SNAP软件,由两位观察者分别使用AK软件提取12个纹理特征参数,采用组内相关系数(ICC)对数据的一致性进行检验,采用独立样本t检验或Mann-Whitney U检验比较两组纹理特征参数,应用ROC曲线分析纹理特征参数的鉴别效能,应用Logistics回归分析联合纹理特征参数进行效能分析,通过Delong检验比较各单一参数模型AUC间的差异及联合模型与各单一参数模型AUC间的差异。结果:Ki67高表达组DCE-MRI纹理特征参数Max intensity_(Ktrans)(1.48(0.99,2.71))、Energy_(Ktrans)(4.76±1.45)、Max intensity_(Kep)(10.01±6.31)、Energy_(Kep)(3.28±2.44)、Entropy_(Kep)(5.28(4.80,5.52))值均大于低表达组纹理特征参数Max intensity_(Ktrans)(0.92(0.53,1.34))、Energy_(Ktrans)(1.90±2.49)、Max intensity_(Kep)(7.30±5.35)、Energy_(Kep)(1.66±2.17)、Entropy_(Kep)(4.65(4.14,5.25)),差异有统计学意义(P<0.05),AUC值分别为0.758、0.826、0.650、0.748、0.716。Ki67两组间有统计学差异的纹理特征参数联合诊断效能更高,AUC值为0.947,灵敏度为91.1%,特异度为88.6%。经Delong检验,Ki67两组间差异有统计学意义的纹理特征参数联合模型的AUC大于单一参数模型的AUC,差异均具有统计学意义(P<0.05)。结论:纹理特征参数Max intensity、Energy、Entropy能有效预测直肠癌Ki67表达状态,纹理特征参数的联合分析可获得更高的鉴别诊断效能。

Objective:To explore the value of dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)texture analysis in predicting the expression of Ki67 in patients with rectal cancer.Methods:A retrospective collection of 91 patients with primary rectal cancer(56 cases in the Ki67 high expression group and 35 cases in the low expression group)who underwent DCE-MRI scan and had the results of Ki67 immunohistochemical detection.The post-processing images of K^(trans),K_(ep),V_(e) are acquired by GenIQ post-processing software and imported into ITK-SNAP software.Two observers use AK software to extract 12 texture feature parameters respectively,applying intra-group correlation coefficient(ICC)to test the consistency of the data,using independent sample t test or Mann-Whitney U test to compare two sets of texture parameters,using ROC curve to analyze the identification power of texture parameters.The logistic regression analysis is used to analyze the efficiency of the combined texture feature parameters,the difference between the AUC of each single parameter model and the difference between the combined model and the AUC of each single parameter model were compared by Delong test.Results:The texture parameters of the Ki67 high expression group,Max intensity_(Ktrans)(1.48(0.99,2.71)),Energy_(Ktrans)(4.76±1.45),Max intensity_(Kep)(10.01±6.31),Energy_(Kep)(3.28±2.44),Entropy_(Kep)(5.28(4.80,5.52))are all larger than the low expression group Max intensity_(Ktrans)(0.92(0.53,1.34)),Energy_(Ktrans)(1.90±2.49),Max intensity_(Kep)(7.30±5.35),Energy_(Kep)(1.66±2.17),Entropy_(Kep)(4.65(4.14,5.25)),the differences were statistically significant(P<0.05),the AUC values are 0.758,0.826,0.650,0.748,0.716 respectively.The combination of P53texture parameters with differences between the two groups has a higher diagnostic efficiency,with an AUC of 0.947,a sensitivity of 91.1%,and a specificity of 88.6%.After Delong test,there is a difference between the two groups of Ki67.The AUC of the texture parameter combined model is greater than that of the single parameter model,and the difference is statistically significant(P<0.05).Conclusions:Texture feature parameters Max intensity,Energy,Entropy can predict the expression of rectal cancer Ki67 effectively,and the combined analysis of texture analysis feature parameters can achieve higher differential diagnosis performance.