TIAN Zhiwei 1WANG Shilong 1MAO Xijin2
作者信息
- 1. Department of Radiology,Binzhou Medical University Hospital,Binzhou 256600,China
- 2. Department of Radiology,Binzhou Medical University Hospital,Binzhou 256600,China||School of Medical Imging,Binzhou Medical University,Yantai 264003,China
- 折叠
摘要
Abstract
Objective:To explore the predictive value of a machine learning model integrating dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)parameters and clinical risk factors for microsatellite instability(MSI)status in rectal cancer.Materials and Methods:A retrospective analysis was conducted on 150 rectal cancer patients treated at Binzhou Medical University Hospital between May 2022 and July 2024,including 27 with MSI and 123 with microsatellite stability(MSS).MRI axial T2WI,apparent diffusion coefficient(ADC)maps,and dynamic contrast-enhanced T1WI(DCE-T1WI)were used to delineate the tumor's largest cross-sectional area as the region of interest(ROI).Radiomic features were extracted and reduced to identify optimal features.Independent clinical predictors and DCE parameters for MSI were selected using multivariate logistic regression.The dataset was split into training and validation sets in a 7∶3 ratio.Nine machine learning algorithms,extreme gradient boosting classifier(XGBoost),logistic regression(LR),light gradient boosting machine classifier(LGBM),random forest(RF),decision tree classifier(DT),Gaussian naive Bayes(GNB),support vector classifier(SVM),multilayer perceptron classifier(MLP),and adaptive boosting classifier(AdaBoost)were employed to construct predictive models.The performance of each model in predicting MSI status was evaluated using receiver operating characteristic(ROC)curves and decision curve analysis(DCA).Additionally,a temporal validation set comprising 30 rectal cancer patients from the same hospital between December 2024 and June 2025 was used to assess model generalizability via ROC analysis.Results:The GNB model demonstrated the most stable performance.The combined model,incorporating independent clinical risk factors,radiomics scores,and DCE perfusion parameters,demonstrated superior predictive performance for MSI status in rectal cancer,with area under the curve(AUC)values of 0.920(95%CI:0.821 to 1.000),0.900(95%CI:0.786 to 1.000),and 0.817(95%CI:0.667 to 0.966)in the training set,validation set,and temporal validation set,respectively.Conclusions:Among the nine machine learning algorithms evaluated,GNB exhibited the best performance in predicting MSI status in rectal cancer,which was further validated using a temporal validation set.Machine learning models incorporating DCE-MRI parameters and clinical risk factors show promising value in predicting MSI status in rectal cancer.关键词
直肠癌/微卫星不稳定性/磁共振成像/动态对比增强/机器学习Key words
rectal cancer/microsatellite instability/magnetic resonance imaging/dynamic contrast-enhanced/machine learning分类
医药卫生
