基于CT增强扫描动脉期图像的纹理分析在肝细胞癌病理分化程度预测中的价值研究  

The Value of Texture Analysis Based on CT Enhanced Arterial Phase Image in Predicting the Pathological Grading of Hepatocellular Carcinoma

作　　者：岑永义 农海洋 李传[3] 梁克 鲍保成 陆玉敏 CEN Yongyi;NONG Haiyang;LI Chuan(Youjiang Medical University for Nationalities,Baise,Guangxi Zhuang Autonomous Region 533000,P.R.China)

机构地区：[1]右江民族医学院,百色533000 [2]广西医科大学第三附属医院放射科,南宁530031 [3]广西壮族自治区人民医院放射科,南宁530016 [4]广西中医药大学第一附属医院放射科,南宁530023

出　　处：《临床放射学杂志》2022年第8期1476-1481,共6页Journal of Clinical Radiology

基　　金：广西壮族自治区中医药局自筹经费科研课题项目(编号:20210681)。

摘　　要：目的探讨基于CT增强扫描动脉期的纹理分析在肝细胞癌(HCC)病理分化程度预测中的价值。方法将病理证实为HCC的患者99例(包括106个病灶,所有病灶术后病理均诊断为HCC)作为研究对象,根据病理结果分为中低分化、高分化两组。采用MaZda软件对所有病例术前CT增强扫描动脉期的原始图像进行纹理特征提取并用MaZda自带的“分类错误概率联合平均相关系数(POE+ACC)”功能筛选出意义最显著的10个纹理特征。选用t检验或Mann-Whitney U检验对这10个纹理特征进行组间比较,使用受试者工作特征曲线(ROC)计算单个纹理特征鉴别HCC病理分化程度的效能,最后用二元Logistic回归模型对差异有统计学意义且曲线下面积(AUC)>0.5的纹理参数行进一步自变量筛选并建立预测模型,采用AUC值、灵敏度、特异度评价该预测模型的预测效果,P<0.05为差异有统计学意义。结果“POE+ACC”筛选出的纹理参数中,共有7个纹理特征在HCC中低分化、高分化组的组间比较中差异有统计学意义(P<0.05),除峰度(Kurtosis)外,其余的6个参数AUC值均>0.5,其范围为0.634~0.805。二元Logistic回归模型显示第90百分位数(Perc.90%)(P=0.004)、低频分量小波系数能量s-3(WavEnLL_s-3)(P=0.020)可作为HCC病理分化程度的独立预测因子,以Perc.90%及WavEnLL_s-3建立的预测模型对HCC分化程度预测的总体准确率为79.2%,其诊断中低分化HCC的AUC值为0.837,灵敏度为89.4%,特异度为62.5%。结论基于CT增强扫描动脉期的纹理分析对初步预测HCC的病理分化程度有一定的价值。Objective To explore the value of texture analysis based on CT enhanced arterial phase in predicting the pathological grade of hepatocellular carcinoma(HCC).Methods 99 patients with HCC confirmed by pathology(including 106 lesions,all lesions were diagnosed as HCC by postoperative pathology)were divided into low differentiation and high differentiation groups according to the pathological results.MaZda software was used to extract texture features from the original images of preoperative CT enhanced arterial phase of all cases,and the“POE+ACC”function of MaZda was used to screen out the 10 most significant texture features.The t test or Mann-Whitney U test were used to compare the 10 texture features between groups,and the efficacy of a single texture feature to identify the pathological grade of HCC was calculated by subject work specific curve(ROC).Finally,the texture parameters with statistical difference and AUC>0.5 were selected by binary logistic regression model,and the prediction model was established.The prediction effect of the prediction model was evaluated by specificity,and the difference was statistically significant(P<0.05).Results Among the texture parameters screened by POE+ACC,there were 7 texture features with statistically significant differences between the low differentiation and high differentiation groups of HCC(P<0.05).Except kurtosis,the AUC values of the other 6 pa-rameters were greater than 0.5,ranging from 0.634 to 0.805.Binary Logistic regression model showed Perc.90%(P=0.004)and WavEnLL_s-3(P=0.020)can be used as an independent predictor of HCC pathological grade,The overall accuracy of the prediction model established by Perc.90%and WavEnLL_s-3 in predicting the degree of differentiation of HCC was 79.2%,the AUC value of low differentiated HCC was 0.837,the sensitivity was 89.4%,and the specificity was 62.5%.Conclusion Texture analysis based on CT enhanced arterial phase is of certain value in predicting the pathological differentiation of HCC.

关 键 词：肝细胞癌 体层摄影术 X线计算机 纹理分析 病理分级 

分 类 号：R735.7[医药卫生—肿瘤] R730.44[医药卫生—临床医学]