摘要
目的 探讨多参数MRI影像组学模型鉴别高级别胶质瘤与原发性中枢神经系统淋巴瘤(primary central nervous system lymphoma, PCNSL)的价值。材料与方法 回顾性分析99例经病理证实为高级别胶质瘤或PCNSL患者的术前常规MRI图像,按7∶3比例随机分为训练组(n=69)及验证组(n=30)。ROI1包括肿瘤核心,ROI2包括肿瘤核心及瘤周水肿区,于轴位对比增强(contrast enhancement, CE)-T1WI、T2液体衰减反转恢复(fluid-attenuated inversion-recovery, FLAIR)图像上进行ROI的勾画。独立样本t检验或Mann-Whitney U检验、皮尔逊相关性分析和最小绝对收缩和选择算法(least absolute shrinkage and selection operator, LASSO)用于特征筛选,并计算每个患者的影像组学评分。采用逻辑回归(logistic regression, LR)算法构建基于CE-T1WI、T2 FLAIR的单一序列和两者联合模型,受试者工作特征(receiver operating characteristic, ROC)曲线来评估预测模型的性能,计算相应模型的曲线下面积(area under the curve, AUC)、准确率、敏感度、特异度。结果 单一序列模型中,CE-T1WI模型预测性能最佳,其在训练集及验证集的AUC值分别为0.952和0.949;提取肿瘤核心特征建立的T2 FLAIR模型优于以全肿瘤区域特征所建模型,其在训练集及验证集的AUC值分别为0.915和0.898;联合模型在训练集及验证集的AUC分别为0.978和0.983。结论 多参数MRI影像组学模型在区分PCNSL和高级别胶质瘤方面具有良好的诊断效能;单一序列模型中,CE-T1WI模型效能最佳,基于不同序列构建的联合模型可以提高模型准确性;肿瘤核心区域特征与肿瘤分类任务更相关。
Objective:To explore the value of multi-parametric MRI-based radiomics models in differentiating primary central nervous system lymphoma(PCNSL)from high-grade glioma.Materials and Methods:All preoperative routine MRI images of 99 patients with high-grade gliomas and primary PCNSLs who confirmed by pathology were collected,and all patients were randomly divided into training(n=69)and testing(n=30)sets at 7:3 ratios.ROI1 included the of core the tumor,ROI2 included the core of tumor and peritumoral edema.Delineated the ROI on the axial contrast enhancement(CE)-T1WI and T2 fluid-attenuated inversion-recovery(FLAIR)images.The independent sample t-test or the Mann-Whitney U test,the Pearson correlation analysis and the least absolute shrinkage and selection operator(LASSO)were used to screen out the features,the radiomics score of each patients were also calculated.We used the logistic regression(LR)algorithm to construct models for CE-T1WI,T2 FLAIR and a combined model as well.The receiver operating characteristic(ROC)was used to evaluate classifier performance,calculating the area under curve(AUC),accuracy,sensitivity,specificity to the corresponding radiomics model.Results:Among single-sequence radiomics models,the CE-T1WI model had the best prediction performance,its AUC values in the training and testing groups were 0.952 and 0.949,respectively.The T2 FLAIR model established by the core features of the tumor is superior to the model based on the whole tumor,its AUC values in the training and testing groups were 0.915 and 0.898,respectively.The AUC values of the combined model in the training and the testing groups were 0.978 and 0.983,respectively.Conclusions:Multi-parametric MRI-based radiomics models had good diagnostic performance in differentiating PCNSL from high-grade glioma,among single-sequence radiomics models,the CE-T1WI model had the best prediction performance,the combined model increased the accuracy of the model,the features of tumor core area are more related to tumor classification.
作者
张少茹
周云舒
张若弟
刘世莉
陈晓华
王卓
陈志强
ZHANG Shaoru;ZHOU Yunshu;ZHANG Ruodi;LIU Shili;CHEN Xiaohua;WANG Zhuo;CHEN Zhiqiang(Clinical Medicine School of Ningxia Medical University,Yinchuan 750003,China;Department of Radiology,the First Affiliated Hospital of Hainan Medical College,Haikou 570102,China;Department of Radiology,General Hospital of Ningxia Medical University,Yinchuan 750004,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2023年第10期53-57,64,共6页
Chinese Journal of Magnetic Resonance Imaging
基金
宁夏回族自治区重点研发计划项目(编号:2019BEG03033)
宁夏回族自治区自然科学基金(编号:2022AAC03472)
教育部春晖项目(编号:Z2012002)。
