基于多序列MRI影像组学预测脑胶质瘤Ki-67表达水平
Prediction of Ki-67 expression level in gliomas based on multi-sequence MRI radiomics
饶定才 1鲁芳 1周丽雯 1岳文军1
作者信息
- 1. 637000 四川南充,川北医学院附属医院放射科
- 折叠
摘要
目的:探讨基于多序列MR图像所构建的影像组学模型在预测脑胶质瘤Ki-67表达水平中的价值.方法:回顾性分析2017年1月—2023年5月经病理检查证实的114例胶质瘤患者的MRI图像和临床特征.根据肿瘤标本检测获得的Ki-67表达水平,将患者分为高表达组(n=54)和低表达组(n=60).将所有患者按7︰3的比例随机分为训练集79例(高表达组37例,低表达组42例)和测试集35例(高表达组17例,低表达组18例).使用ITK-SNAP软件,在CE-T1WI图像上沿胶质瘤边缘逐层勾画ROI,获取肿瘤全域容积ROI(VOI).然后使用"Pyradiomics"软件包,分别自T1WI、CE-T1WI、T2WI和T2-FLAIR图像上提取肿瘤VOI的影像组学特征.采用t检验和递归特征消除法进行特征筛选,利用随机森林方法建立预测Ki-67表达水平的临床-影像模型、影像组学模型及联合模型.采用ROC曲线下面积(AUC)和DeLong检验来评估和比较上述模型的效能.使用校准曲线和Hosmer-Lemeshow检验分析模型的拟合优度,并通过决策曲线分析(DCA)评估各个模型的临床应用价值.结果:临床-影像模型、影像组学模型和联合模型在训练集和测试集中的AUC分别为0.791(95％CI:0.685～0.874)和0.716(95％CI:0.538～0.855)、0.907(95％CI:0.820～0.961)和0.866(95％CI:0.708～0.957)、0.964(95％CI:0.896～0.993)和0.908(95％CI:0.760～0.979),以联合模型的的AUC最大.Delong检验结果显示,联合模型的预测效能优于临床影像模型(训练集P<0.001;测试集P=0.018),但与影像组学模型之间的差异无统计学意义(P>0.05).Hosmer-Lemeshow检验结果显示联合模型校准曲线的拟合度良好(P>0.05).DCA结果显示,当风险阈值为0.11～0.89时,联合模型的临床净效益高于影像组学模型.结论:基于临床、影像特征及多序列MRI影像组学特征构建的联合模型对术前预测胶质瘤Ki-67表达水平具有一定价值.
Abstract
Objective:The purpose of this study was to explore the value of radiomics model based on multi-sequence MRI images in predicting Ki-67 expression level in gliomas.Methods:MRI images and cliniccal features of 114 glioma patients confirmed by pathology from January 2017 to May 2023 were analyzed retrospectively.According to the Ki-67 expression level obtained from tumor speci-men detection,patients were divided into high expression group (n=54)and low expression group (n=60).All patients were randomly divided into training set of 79 cases (37 patients in the high ex-pression group and 42 patients in the low expression group)and test set of 35 cases (17 patients in the high expression group and 18 patients in the low expression group)at a ratio of 7︰3 .ITK-SNAP soft-ware was used to delineate the regions of interest (ROI)along the edge of glioma on CE-T1WI image to obtain the volume ROI (VOI)of the tumor,and radiomics features were extracted from tumor VOI on T1WI,CE-T1WI,T2WI,and T2-FLAIR images respectively with the"Pyradiomics"software pac-kage.Feature selection was completed by t-test and recursive feature elimination.Clinic-radiological model,radiomics model and combined model were established by random forest method to predict Ki-67 expression level.The performance of the above models were evaluated and compared using the area under the receiver operating characteristic (ROC)curve (AUC)and DeLong test.The goodness of fit of the models was analyzed using calibration curves and the Hosmer-Lemeshow test,and the clinical application value of each model was assessed by decision curve analysis (DCA).Results:The AUC values of clinic-radiological model,radiomics model,and combined model in the training set and test set were 0.791 (95％CI:0.685～0.874)and 0.716 (95％CI:0.538～0.855),0.907 (95％CI:0.820～0.961)and 0.866 (95％CI:0.708～0.957),0.964 (95％CI:0.896～0.993)and 0.908 (95％CI:0.760～0.979),respectively;the AUC of the combined model was of the highest.Delong test showed that the prediction performance of the combined model was better than the clinic-radiological model (training set:P<0.001;test set:P=0.018),but there was no statistically significant difference in AUC between the combined model and the radiomics model (P>0.05).The results of the Hosmer-Lemeshow test demonstrated that the calibration curve of combined models had a satisfactory degree of fit (P>0.05).The DCA results indicated that the clinical net benefit of the combined model was higher than those of the radiomics model when the risk threshold was 0.11 to 0.89.Conclusion:The combined model based on clinic-radiological features and multi-sequence MRI radiomics features has certain value in predicting the Ki-67 expression level in gliomas before surgery.
关键词
胶质瘤/磁共振成像/影像组学/Ki-67Key words
Gliomas/Magnetic resonance imaging/Radiomics/Ki-67引用本文复制引用
出版年
2024
NETL
NSTL
万方数据