磁共振成像2025，Vol.16Issue(8)：41-49,72,10.DOI:10.12015/issn.1674-8034.2025.08.007
基于临床-多模态磁共振成像影像组学的胶质母细胞瘤与原发性中枢神经系统淋巴瘤无创鉴别模型构建及验证
Construction and validation of a non-invasive differentiation model for glioblastoma and primary central nervous system lymphoma based on clinical-multimodal magnetic resonance imaging radiomics
摘要
Abstract
Objective:To overcome the limitations of conventional imaging in differentiating glioblastoma(GBM)from primary central nervous system lymphoma(PCNSL),we propose and validate a clinically integrated radiomics model for the preoperative,non-invasive stratification of these two oncological entities.Materials and Methods:A retrospective cohort of 173 patients with intracranial masses(118 GBM,55 PCNSL),confirmed by histopathology or diagnostic radiotherapy,was randomly divided into training(n=121)and validation(n=52)sets in a 7∶3 ratio.Preoperative clinical parameters(serological indices,imaging manifestations)and multimodal MRI sequences[CE-T1WI,T2-FLAIR,DWI(b=1000 s/mm2),and ADC]were acquired.Tumor core regions(excluding peritumoral edema)were delineated as regions of interest(ROIs).Following Z-score normalization,key features were selected using the Mann-Whitney U test,Spearman correlation analysis,and the least absolute shrinkage and selection operator(LASSO)algorithm.An XGBoost classifier with 10-fold cross-validation was employed for model construction.A comparative analysis of five models was performed:the clinical model,four single-modality radiomics models,the multimodal radiomics model,and the integrated clinical-radiomics model.The diagnostic performance was evaluated using receiver operating characteristic(ROC)curves,with the area under the curve(AUC),sensitivity,specificity,and accuracy calculated.The statistical validation included the DeLong test for AUC comparison,calibration curve assessment,and decision curve analysis(DCA)to quantify clinical utility.Results:The clinical model demonstrated AUC values of 0.83(95%CI:0.76 to 0.90)in the training set and 0.74(95%CI:0.61 to 0.87)in the validation set.Among radiomics models,the multimodal radiomics model(T1+ADC+T2+DWI)achieved optimal performance with training/validation AUCs of 0.93(95%CI:0.88 to 0.98)/0.84(95%CI:0.72 to 0.96).The integrated clinical-radiomics model demonstrated superior diagnostic performance,achieving a training AUC of 0.94(95%CI:0.90 to 0.98)(accuracy 90.2%,sensitivity 96.7%)and a validation AUC of 0.85(95%CI:0.74 to 0.96)(accuracy 88.6%,sensitivity 83.3%).This combined model significantly outperformed individual models in predictive accuracy(DeLong test,P<0.05)and clinical net benefit across threshold probability ranges(decision curve analysis).Conclusions:The combined model,constructed by integrating clinical features and multimodal radiomics,can non-invasively and stably distinguish GBM from PCNSL,providing reliable references for the precise preoperative diagnosis of patients.It helps reduce the need for invasive tests and optimizes the clinical decision-making process.关键词
胶质母细胞瘤/原发性中枢神经系统淋巴瘤/临床特征/影像组学/多模态磁共振成像/鉴别诊断Key words
glioblastoma/primary central nervous system lymphoma/clinical features/radiomics/multimodal magnetic resonance imaging/differential diagnosis分类
医药卫生引用本文复制引用
宋婷婷,洪士强,祝贺,郑蕾,吴昌顺,冯虹..基于临床-多模态磁共振成像影像组学的胶质母细胞瘤与原发性中枢神经系统淋巴瘤无创鉴别模型构建及验证[J].磁共振成像,2025,16(8):41-49,72,10.基金项目
National Natural Science Foundation of China(No.81672379,81473483). 国家自然科学基金项目(编号:81672379、81473483) （No.81672379,81473483）
