摘要
Abstract
Objective To construct a prediction model for pathological differentiation grade of intrahepatic cholangiocarcinoma(ICC)based on contrast-enhanced magnetic resonance imaging(MRI)radiomics combined with machine learning algorithms and clinical features,and evaluate its efficacy.Methods The clinical data of 100 ICC patients admitted in the First Affiliated Hospital of Wenzhou Medical University from January 2013 to December 2022 were retrospectively collected.Based on the postoperative pathological differentiation grade of the tumor,the patients were divided into the poorly differentiated group(n=22)and the non-poorly differentiated group(n=78).LASSO regression screened clinical features associated with the differentiation grade of ICC,followed by multivariate Logistic regression to identify independent predictors for the establishment of the model.Radiomics features were extracted from preoperative contrast-enhanced MRI sequences,and multiple machine learning classifiers were trained and validated via 10-fold cross-validation to establish the single-sequence prediction model and assess its efficacy.The radiomics features of two sequences with the best efficacy were integrated into the double-sequence prediction model and further combined with clinical features to construct the radiomics-clinical prediction model.Receiver operating characteristic(ROC)curves,calibration curves,and decision curves were used to evaluate the diagnostic performance and clinical utility of the model.Results LASSO regression revealed a best efficacy in predicting the poor differentiation grade of ICC when the model contained clinical features including ill-demarcated margins of the tumor,intratumoral arterial traversal signs,age>60 years and alanine aminotransferase(ALT)>40 U/L.Multivariate Logistic regression demonstrated that heterogeneity of intratumoral arteries(OR=4.437,95%CI 1.551 to 14.282,P=0.008)and age>60 years(OR=0.212,95%CI 0.069 to 0.594,P=0.004)were independent prediction factors of the poor differentiation grade of ICC.Comparison of machine learning classifiers demonstrated the best performance in the double-sequence prediction model that included radiomics features of the arterial phase and DWI sequences(validation set AUC=0.837).The radiomics-clinical prediction model had improved predictive accuracy:training set area under the curve(AUC)=0.944(sensitivity=0.941,specificity=0.875),validation set AUC=0.908(sensitivity=0.750,specificity=0.852),and test set AUC=0.867(sensitivity=0.800,specificity=0.887).Decision curve analysis(DCA)confirmed net clinical benefit in the threshold probability range of 0.10-0.75.Conclusion The MRI radiomics-clinical prediction model constructed in this study has good efficacy in predicting the differentiation degree of ICC preoperatively,which may guide individualized treatment in clinical practice.关键词
肝内胆管癌/病理分化程度/影像组学/磁共振成像/机器学习Key words
intrahepatic cholangiocarcinoma/pathological differentiation degree/radiomics/magnetic resonance imaging/machine learning分类
医药卫生
