【摘要】 目的 构建预测肝细胞癌患者发生微血管侵犯（MVI）的临床-MRI影像组学列线图模型，探讨其预测价值。方法 2019年6月—2020年11月河南省人民医院诊治肝细胞癌患者118例，随机分为训练集83例和验证集35例。比较训练集与验证集血清谷丙转氨酶（ALT）、谷草转氨酶（AST）、甲胎蛋白（AFP）、血小板与淋巴细胞比值（PLR）、碱性磷酸酶与淋巴细胞比值（ALR）、（碱性磷酸酶+γ-谷氨酰转移酶）/淋巴细胞计数（AGLR）及MVI发生率。采用多因素logistic回归分析肝细胞癌患者发生MVI的临床影响因素。118例患者均行MRI检查，应用ITK-SNAP软件在T₂压脂快速自旋回波序列（T₂WFS）、弥散加权成像（DWI）图像上勾画感兴趣区并将数据进行归一化处理，提取影像组学特征，采用lasso回归筛选最优子集并计算放射组学评分（Rad-Score）。根据临床影响因素及MRI影像组学Rad-Score构建预测肝细胞癌患者发生MVI的临床-MRI影像组学列线图模型。绘制ROC曲线，评估临床-MRI影像组学列线图模型预测肝细胞癌患者发生MVI的效能；绘制校准曲线，评估列线图模型的预测效率和观测概率；绘制决策曲线，评估列线图模型的净收益；采用Hosmer-Lemeshow检验分析列线图模型的拟合度。结果 训练集血清ALT、AST、AFP水平及PLR、ALR、AGLR、MVI发生率与验证集比较差异均无统计学意义（Z=-0.961～0.649,χ²=0.103,P均>0.05）。年龄（OR=1.087,95%CI:1.032～1.156,P=0.003）、ALR（OR=1.790,95%CI:1.169～2.991,P=0.015）、AGLR（OR=1.022,95%CI:1.011～1.034,P=0.001）是肝细胞癌患者发生MVI的临床影响因素。118例患者轴位T₂WFS和DWI图像中定量提取1 926个影像组学特征，包括原始图像中的104个原始特征和滤波器变换处理后的1 822个特征，生成T₂WFS&DWI特征集，lasso回归筛选出由17个特征组成的最优子集并计算Rad-Score,预测肝细胞癌患者发生MVI的Rad-Score阈值为0.50。根据临床影响因素及MRI影像组学Rad-Score构建预测肝细胞癌患者发生MVI的临床-MRI影像组学列线图模型。在训练集中，临床-MRI影像组学列线图模型预测肝细胞癌患者发生MVI的AUC为0.924（95%CI:0.868～0.980,P=0.041）,灵敏度为76.7%,特异度为95.1%;在验证集中，临床-MRI影像组学列线图模型预测肝细胞癌患者发生MVI的AUC为0.853（95%CI:0.728～0.978,P=0.028）,灵敏度为71.0%,特异度为77.8%。校准曲线结果显示，临床-MRI影像组学列线图模型预测肝细胞癌患者发生MVI的结果与实际观测结果一致性良好。决策曲线结果显示，临床-MRI影像组学列线图模型在较多数阈值概率范围内净收益良好。Hosmer-Lemeshow检验结果显示临床-MRI影像组学列线图模型的拟合度高（χ²=7.530,P=0.481）。结论 基于MRI影像组学联合临床特征构建的临床-MRI影像组学列线图模型对肝细胞癌患者发生MVI有一定预测价值。更多还原

【Abstract】 Objective To construct a clinical-MRI radiomics nomogram model for predicting microvascular invasion（MVI） in patients with hepatocellular carcinoma（HCC）,and to explore its predictive value.Methods Totally 118 HCC patients were diagnosed and treated in Henan Provincial People’ s Hospital from June 2019 to November 2020,and were randomly divided into the training set（n=83） and the validation set（n=35）.The serum alanine transaminase（ALT）,aspartate transaminase（AST）,alpha-fetoprotein（AFP）,platelet to lymphocyte ratio（PLR）,alkaline phosphatase to lymphocyte ratio（ALR）,ratio of alkaline phosphatase+γ-glutamyl transpeptidase to lymphocyte count（AGLR）,and MVI incidence were compared between the training set and the validation set.Multivariate logistic regression was used to analyze the clinical influencing factors of MVI in HCC patients.All 118 patients underwent MRI examination.ITK-SNAP software was used to draw the region of interest on T₂-weighted fat suppression fast spin echo sequence（T₂ WFS） and diffusion weighted imaging（DWI） images,and the data were normalized.The radiomics features were extracted,and the lasso regression was used to screen the optimal subset and calculate the radiomics score（Rad-Score）.According to clinical influencing factors and MRI radiomics Rad-Score,a clinical-MRI radiomics nomogram model was constructed to predict MVI in HCC patients.ROC curves were plotted to evaluate the efficiency of clinical-MRI radiomics nomogram model on predicting MVI in HCC patients.The calibration curves were drawn to evaluate the prediction efficiency and observation probability of the nomogram model.The decision curves were drawn to evaluate the net benefit of the nomogram model.The Hosmer-Lemeshow test was done to analyze the fitting degree of the nomogram model.Results There were no significant differences in the serum ALT,AST,AFP,PLR,ALR,AGLR and MVI incidence between the training set and the validation set（Z=-0.961 to 0.649,χ²=0,103;all P values >0.05）.Age（OR=1.087,95%CI:1.032-1.156,P=0.003）,ALR（OR=1.790,95% CI:1.169-2.991,P=0.015）,and AGLR（OR=1.022,95%CI:1.011-1.034,P=0.001） were the clinical influencing factors of MVI in HCC patients.A total of 1 926radiomics features were quantitatively extracted from axial T₂ WFS and DWI images of 118 patients,including 104 original features from the original images and 1 822 features after filter transformation processing.The T₂ WFS&DWI feature set was generated.The optimal subset composed of 17 features was screened with lasso regression and the Rad-Score was calculated.The Rad-Score threshold was 0.50.Based on clinical influencing factors and MRI radiomics Rad-Score,a clinical-MRI radiomics nomogram model was constructed.In the training set,the AUC of clinical-MRI radiomics nomogram model for predicting MVI in HCC patients was 0.924（95%CI:0.868-0.980,P=0.041）,with a sensitivity of76.7% and a specificity of 95.1%.In the validation set,the AUC of the clinical-MRI radiomics nomogram model for predicting MVI in HCC patients was 0.853（95%CI:0.728-0.978,P=0.028）,with a sensitivity of 71.0% and a specificity of 77.8%.The results of calibration curves showed that the clinical-MRI radiomics nomogram model for predicting MVI in HCC patients had a good consistency with the actual observation results.The results of the decision curves showed that the clinical-MRI radiomics nomogram model achieved a good net benefit in the range of most threshold probabilities.The results of Hosmer-Lemeshow test showed that the clinical-MRI radiomics nomogram model had a high fitting degree（χ²=7.5 30,P=0.481）.Conclusion The clinical-MRI radiomics nomogram model based on MRI radiomics combined with clinical features has a certain predictive value for MVI in HCC patients.更多还原