【摘要】 目的 探讨构建基于规范化增强MRI影像组学列线图术前预测子宫内膜癌淋巴血管侵犯（LVSI）的价值。方法 选取157例子宫内膜癌患者术前盆腔MRI检查及临床资料，按照7∶3的比例，随机分为训练集和验证集。从CE-T₁WI、T₂WI、DWI和ADC图中提取影像组学特征。利用最小相关最大冗余性（mRMR）算法和递归特征消除（RFE）选择特征，KNN（K-Nearest Neighbor）算法构建组学模型。采用多变量logistic回归分析，结合Rad-score和临床危险因素，构建列线图，比较影像组学特征、临床危险因素和列线图对LVSI的预测效能。结果 1781个影像组学特征经mRMR和RFE算法筛选出12个影像组学特征，并计算Rad-score。单因素和多因素logistic回归分析发现CA125、肿瘤级别以及Radscore为LVSI的危险因素。训练集和验证集中影像组学模型的预测性能为0.83（95%CI:0.76～0.91）和0.78（95%CI:0.65～0.91）；临床模型的预测性能分别为0.83（95%CI:0.72～0.93）和0.75（95%CI:0.55～0.95）；列线图预测性能分别为0.90（95%CI:0.84～0.97）和0.86（95%CI:0.72～1.00）。列线图对LVSI具有较高的预测效能，并且净分类指数（NRI）和总综合鉴别指数（IDI）证实列线图优于临床模型。结论 基于规范化增强MRI影像组学列线图可用于术前预测EC LVSI。更多还原

【Abstract】 Objective To construct normalized enhancement radiomics based MRI radiomics nomogram and to evaluate its value in predicting lymphatic vascular space invasion （LVSI） of endometrial carcinoma （EC） before operation.Methods 157patients with endometrial carcinoma were retrospectively analyzed in pelvic MRI examination and clinical data before operation.They were randomly divided into training group and verification group according to the ratio of 7∶3.Radiologic features were extracted from CE-T₁WI,T₂WI,DWI and ADC images.The minimum redundancy maximum relevance （mRMR） and recursive feature elimination （RFE） were used to select features,and KNN （K-Nearest Neighbor） was used to build an radiomics model.Multivariate logistic regression analysis was used to construct nomograms based on the Rad-score and clinical risk factors,and to compare the predictive efficacy of radiomics,clinical risk factors and nomograms on LVSI.Results 12 of 1781 radiomics features were selected by mRMR and RFE,then Rad-score was calculated.Univariate and multivariate logistic regression analysis showed that CA125,tumor grade and Rad-score were risk factors for LVSI.In the training group and validation group,the predictive performance of the radiomics was 0.83 （95%CI:0.76～0.91） and 0.78 （95%CI:0.65～0.91）;The predictive performance of the clinical model was 0.83 （95%CI:0.72～0.93） and 0.75 （95%CI:0.55-0.95） respectively;The predicted performance of nomogram was 0.90 （95%CI:0.84-0.97） and 0.86 （95%CI:0.72～1.00） respectively.The nomogram had a high prediction efficiency for LVSI,and the net reclassification index （NRI） and the integrated discrimination improvement （IDI） confirmed that the nomogram was superior to the clinical model.Conclusion The radiomics nomogram based on normalized enhancement radiomics can be used to predict EC LVSI before operation.更多还原