基于钆塞酸二钠增强MRI影像组学列线图模型术前预测结直肠癌肝转移瘤微血管侵犯  

Preoperative Prediction of Microvascular Invasion in Colorectal Cancer Liver Metastases Based on Gd-EOB-DTPA Enhanced MRI Radiomics Nomogram Model

作　　者：卞雪莲 张培培 孙琦 王咪 董瀚韵 戴晓晓 吴永友[3] 张力元[4] 范国华[1] 陈光强[1] BIAN Xuelian;ZHANG Peipei;SUN Qi(Department of Radiology,The Second Affiliated Hospital of Soochow University,Suzhou,Jiangsu Province 215004,P.R.China)

机构地区：[1]苏州大学附属第二医院放射科,215004 [2]苏州大学附属第二医院病理科,215004 [3]苏州大学附属第二医院普外科,215004 [4]苏州大学附属第二医院放疗科,215004 [5]泗洪医院放射科,宿迁223900

出　　处：《临床放射学杂志》2024年第7期1118-1124,共7页Journal of Clinical Radiology

基　　金：江苏省重点研发计划专项项目(编号:BE2021652);中核医疗产业有限公司“技术创新”专项项目(编号:ZHYLTD2021001)。

摘　　要：目的基于钆塞酸二钠(Gd-EOB-DTPA)增强MRI肝胆期影像组学建立并验证术前预测结直肠癌肝转移瘤(CRLM)微血管侵犯(MVI)的列线图模型。方法回顾性连续搜集行Gd-EOB-DTPA增强MRI检查且经病理诊断为CRLM的患者共63例,其中MVI阳性22例、MVI阴性41例。应用单因素及多因素二元Logistic回归分析性别、年龄、肿瘤指标等临床指标及MRI征象,筛选出对预测CRLM的MVI有统计学意义的临床及MRI征象,建立临床模型。从Gd-EOB-DTPA增强MRI肝胆期提取并筛选影像组学特征,最后使用逻辑回归(LR)、线性判别分析(LDA)和支持向量机(SVM)三种分类器构建影像组学模型并使用五折交叉验证。此外,还构建了结合临床模型和最佳组学模型的列线图模型。通过受试者工作特征(ROC)曲线、校准曲线及决策曲线分析(DCA)进行效能评估。结果单因素及多因素结果显示T_(2)WI信号不均匀、肝胆期瘤周低信号是CRLM病灶MVI的独立预测因子。最终LR分类器构建的影像组学模型在训练组和测试组中均有着相对较高的预测效能。临床模型联合影像组学模型所构建的列线图模型具有最佳的预测效能,训练组和测试组中的曲线下面积(AUC)分别为0.970(95%CI:0.911~1.000)、0.917(95%CI:0.798~1.000)。校准曲线表明列线图模型的拟合优度良好,DCA表明列线图模型在预测CRLM的MVI时具有较高的临床净获益率。结论基于Gd-EOB-DTPA增强MRI肝胆期影像组学与MRI征象(T_(2)WI信号不均匀、肝胆期瘤周低信号)相结合的列线图模型可作为术前检测CRLM病灶MVI的辅助工具。Objective Development and validation of a nomogram model for preoperative prediction of microvascular invasion(MVI)in colorectal cancer liver metastases(CRLM)based on Gd-EOB-DTPA enhanced MRI hepatobiliary phase radiomics features.Methods A total of 63 patients who underwent Gd-EOB-DTPA enhanced MRI and were pathologically diagnosed with CRLM were retrospectively and consecutively collected,of which 22 were MVI-positive and 41 were MVI-negative.Univariate and multivariate binary logistic regression analysis were applied to analyze the clinical indicators such as gender,age,tumor indexes and MRI features to screen out the clinical and MRI features that were statistically significant in predicting the MVI of patients with CRLM and to establish a clinical model.Radiomics features were extracted and screened from the hepatobiliary phase of Gd-EOB-DTPA enhanced MRI.Finally,three classifiers,logistic regression(LR),linear discriminant analysis(LDA)and support vector machine(SVM),were used to construct the radiomics models and using fivefold cross validation.In addition,a nomogram model combining the clinical model and the best radiomics model was constructed.Efficacy was evaluated by receiver operating characteristic(ROC)curves,calibration curves and decision curve analysis(DCA).Results The univariate and multivariate results showed that T_(2)WI signal inhomogeneity and peri-tumor low signal in the hepatobiliary stage were independent predictors of MVI in CRLM patients.The radiomics model constructed by the final LR classifier had relatively high predictive efficacy in both the training and test groups.The nomogram model constructed by the clinical model combined with the radiomics model had the best predictive efficacy,with AUCs of 0.970(95%CI:0.911-1.000)and 0.917(95%CI:0.798-1.000)in the training and test groups,respectively.The calibration curves demonstrated the goodness of fit of the nomogram model,and the DCA indicated that the nomogram model had a high net clinical benefit rate in predicting MVI in patients wit

关 键 词：微血管侵犯 结直肠癌肝转移瘤 影像组学 钆塞酸二钠 

分 类 号：R735.34[医药卫生—肿瘤] R445.2[医药卫生—临床医学]