摘要
目的 探讨基于CT影像组学模型鉴别消化系统来源肝转移瘤的价值.方法 回顾性收集多中心200例经病理证实的消化系统肝转移瘤患者,包括结直肠癌80例、胃癌55例和胰腺癌65例.基于增强CT门静脉图像提取并筛选组学特征,使用K近邻(KNN)、回归树(RT)、随机森林(RF)、高斯朴素贝叶斯(NB)、支持向量机(SVM)构建三分类模型,准确度及宏观平均AUC评估模型效能;进一步使用逻辑回归(LR)构建二分类(结直肠癌vs胃癌、结直肠癌vs胰腺癌、胃癌vs胰腺癌)模型,通过受试者工作特征曲线下面积(AUC)、校准曲线、临床决策曲线(DCA)量化模型有效性.结果 三分类模型中验证集的准确度为0.91、0.44、0.48、0.62及0.53,宏观平均AUC为0.95、0.47、0.54、0.68及0.61;二分类验证集中AUC为0.82、0.80及0.78;校准曲线显示训练集与验证集之间有较好一致性;DCA曲线展示临床适用度好.结论 基于肝脏增强CT影像组学模型在鉴别消化系统肝转移瘤原发来源中发挥很好的作用.
Abstract
Objective To investigate the value of CT radiomics model in identifying liver metastases from digestive system.Methods A total of 200 patients with pathologically confirmed liver metastases from the digestive system were retrospectively collected,including 80 cases of colorectal cancer,55 cases of gastric cancer and 65 cases of pancreatic cancer.Based on enhanced CT portal vein image extraction and screening omics features,K-nearest neighbors(KNN),regression tree(RT),random forest(RF),Gaussian naïve Bayes(NB)and support vector machine(SVM)were used to construct a three-class model to evaluate the efficiency of the model with accuracy and macro-averaged AUC.Calibration curves,clinical decision curves(DCA)quantify model validity.Results The accuracy of the validation set of three-class model was 0.91,0.44,0.48,0.62 and 0.53,the macro-averaged AUC was 0.95,0.47,0.54,0.68 and 0.61,the AUC of the validation set of the two-class was 0.82,0.80 and 0.78,the calibration curve showed good consistency between the training set and the validation set,and the DCA curve showed good clinical applicability.Conclusion Liver-enhanced CT radiomics model plays a good role in identifying the primary source of liver metastases in the digestive system.
维普
万方数据