MRI纹理特征与肝细胞肝癌Ki-67表达的相关性

Correlation between MRI texture features and Ki-67 expression in HCC

目的:探讨ADC图、T2WI、DWI的纹理特征诊断肝细胞肝癌(HCC)Ki-67标记指数高低的价值。方法:搜集本院经病理确诊的HCC患者57例并将其分为Ki-67>20%、Ki-67≤20%两组。通过MaZda软件手工勾画兴趣区(ROI)并分别提取ADC图、T2WI、DWI纹理特征,随后采用Fisher系数、分类错误概率联合平均相关系数(POE+ACC)、交互信息(MI)的方法分别选择10种最佳纹理特征。纹理特征的组间比较采用t检验或Mann-Whitney U检验,描绘受试者工作特征(ROC)曲线并计算曲线下面积(AUC)值,并将每个序列组间比较有统计学差异的纹理特征纳入二元logistic回归模型进行自变量筛选,并建立预测模型。应用灵敏度、特异度、AUC值来评估预测模型的分类性能。结果:对于区别HCC的Ki-67标记指数高低,t检验和U检验结果显示ADC图组间差异有统计学意义的纹理特征20个;T2WI序列共选择出组间差异有统计学意义的纹理特征10个;组间比较结果显示DWI纹理特征在Ki-67标记指数高、低组间差异无统计学意义。二元logistic回归显示ADC图中的S(5,0)和方差以及T2WI中的S(5,0)和均值、高频对角分量小波系数能量s-5是Ki-67>20%的独立预测因素。S(5,0)和方差、高频对角分量小波系数能量s-5数值越大,S(5,0)和均值的数值越小,患者Ki-67>20%的风险越高。结合ADC图中的S(5,0)和方差以及T2WI中的高频对角分量小波系数能量s-5,建立HCC Ki-67表达程度的预测模型,AUC值为0.795,敏感度为61.5%,特异度为90.3%。结论:利用MRI影像组学评估HCC的Ki-67表达,其影像组学预测模型具有较高的诊断效能。

Objective:To investigate the diagnostic value of texture analysis on ADC,DWI,and T2WI sequence in the Ki-67 marker index of hepatocellular carcinoma(HCC).Methods:A total of 57 patients who underwent pathological examination were retrospectively enrolled.All patients were divided into the Ki-67>20%and Ki-67≤20%groups.ROI was manually sketched by MaZda software,and texture features of DWI,T2WI,and ADC image were extracted.The texture features were selected by Fisher,POE+ACC and MI.The inter-group comparison was performed for t-test or Mann-Whitney U-test,and the receiver operating characteristic curve was used to calculate area under curve(AUC)value.The selected texture features between each sequence group were incorporated into the binary logistic regression model to screen independent variables and establish a prediction model.Sensitivity,specificity,and AUC were used to evaluate the classification performance of the prediction model.Results:For the Ki-67 marker index of hepatocellular carcinoma,the t-test and U test showed that 20 texture features with statistical significance were selected by ADC,and 10 texture features with statistical significance were selected by T2WI.The results of comparisons between the two groups showed that there was no significant difference of DWI texture features.Bivariate logistic regression shows that S(5,0)sum of variance in ADC,the sum of average,and energy of wavelet transform coefficients in subband HH with subsampling factor s-5 in T2WI were the independent predictors of Ki-67>20%.The predictive model of Ki-67 expression in hepatocellular carcinoma was established by combining S(5,0)sum of variance in ADC with the energy of wavelet transform coefficients in subband HH with subsampling factor s-5 in T2WI.The AUC,sensitivity and specificity were 0.795,90.3%,and 61.5%,respectively.Conclusion:The prediction model based on the MRI could be used to assess the expression of Ki-67 in HCC.