中国中西医结合影像学杂志2025，Vol.23Issue(4)：430-435,6.DOI:10.3969/j.issn.1672-0512.2025.04.006
基于多参数影像组学术前预测原发性中枢神经系统淋巴瘤Ki-67表达的价值
Preoperative prediction of multiparametric radiomics for Ki-67 expression level in primary central nervous system lymphoma
摘要
Abstract
Objective:To explore the value of constructing a machine learning model based on multiparametric radiomics for preoperative identification of Ki-67 expression level in primary central nervous system lymphoma(PCNSL).Methods:A retrospective analysis was constructed on 203 PCNSL patients,with 162 patients from one center serving as the training set and 41 patients from another center as the external validation set.Manual segmentation of lesions on T1WI-CE,DWI,and T2 FLAIR images was performed to extract the optimal radiomics features.Four machine learning algorithms were used to construct radiomic models based on the optimal radiomics features.Independent clinical predictors were identified through univariate and multivariate logistic regression analysis.A combined model incorporating these predictors and the radiomics score based on the optimal radiomic model was developed to predict Ki-67 expression level,and the predictive performance of the combined model was evaluated.Results:There were 203 cases,including 89 cases with Ki-67 high expression and 114 cases with low expression.Tumor maximum diameter,peritumoral edema,maximum standardized uptake value(SUVmax)on 18F-FDG PET/CT were independent predictors of Ki-67 expression level.Ultimately,19 radiomics features were screened out,and four machine learning algorithms were used to construct the radiomics model,among which,the gradient boosting machine(GBM)model had a high AUC value(0.88 for the training set and 0.83 for the external validation set).The Random Forest(RF)model followed,with an AUC value of 0.86 for the training set and 0.79 for the external validation set.The GBM model had higher sensitivity and recall rate,while the RF model had a higher accuracy.The combined GBM model had the best predictive performance,with the highest AUC values for the training set and external validation set being 0.92(95%CI 0.88-0.96)and 0.88(95%CI 0.78-0.97),respectively.Calibration curves indicated good calibration of the combined GBM model.Decision curve analysis showed that the combined GBM model had a higher overall net benefit.Conclusion:A multiparametric radiomics model combined with clinical features can effectively stratify the preoperative Ki-67 expression leveL in PCNSL.关键词
机器学习/影像组学/原发性中枢神经系统淋巴瘤/Ki-67/磁共振成像Key words
Machine learning/Radiomics/Primary central nervous system lymphoma/Ki-67/Magnetic resonance imaging引用本文复制引用
李勰,李辉虎,李繁,万运..基于多参数影像组学术前预测原发性中枢神经系统淋巴瘤Ki-67表达的价值[J].中国中西医结合影像学杂志,2025,23(4):430-435,6.基金项目
茂名市科技计划项目(220402114550456). （220402114550456）
