预测胶质瘤复发和术后生存期的MRI影像组学初步研究  

A preliminary study on predicting glioblastoma recurrence and postoperative survival time through MRI imaging radiomics

作　　者：翟晓阳 任进发 程思佳 毛珂 董亚宁 韩东明[1] ZHAI Xiaoyang;REN Jinfa;CHENG Sijia;MAO Ke;DONG Yaning;HAN Dongming(Department of MRI,the First Affiliated Hospital of Xinxiang University,Xinxiang 453100,China)

机构地区：[1]新乡医学院第一附属医院核磁共振科,新乡453100

出　　处：《磁共振成像》2023年第12期26-32,共7页Chinese Journal of Magnetic Resonance Imaging

摘　　要：目的基于术前MRI影像组特征和临床特征建立早期预测胶质瘤术后复发和生存期的评估模型。材料与方法回顾性分析了120例胶质瘤患者的MRI图像和临床资料,提取了瘤周水肿区域和瘤内增强区域的影像组学特征。使用卡方检验或Fisher's精确检验来比较复发与未复发组之间变量之间的差异,使用t或U检验来检验连续型变量之间的差异。使用t检验、Spearman相关性分析和最小绝对收缩和选择算子(least absolute shrinkage and selection operator regression,LASSO)回归对特征进行降维。建立了瘤内、瘤内+瘤周水肿和融合模型的三种预测模型。使用诺模图展示生存期的预测情况,Kaplan-Meier(KM)图展示不同分组之间生存情况。结果在复发组与未复发组之间异柠檬酸脱氢酶(isocitrate dehydrogenase,IDH)状态和影像组学评分(Rad-score)差异有统计学意义,P值分别为0.04和<0.001。最终纳入15个影像组学特征,训练集中三种模型的曲线下面积(area under the curve,AUC)分别为0.905、0.925和0.923,在测试集中的AUC分别为0.859、0.866和0897,融合模型达到最优效果。KM分析显示在训练集和测试集中不同分组患者的生存时间无明显差异。结论基于MRI的影像组学预测胶质瘤患者术后复发有较好的效果,并且可以初步评估术后生存期。Objective:To develop an evaluation model for predicting early postoperative recurrence and evaluating the prognosis of glioma patients using preoperative MRI radiomics and clinical features.Materials and Methods:The MRI images and clinical data of 120 glioma patients were analyzed retrospectively to extract the imaging omics characteristics of the peritumoral edema areas and intratumoral enhancement areas.To compare the variables between the recurrence and non-recurrence groups,we utilized either Chi-square tests or Fisher's exact tests.Furthermore,differences between continuous variables were examined through t-tests or U tests.For dimensionality reduction of the features,we employed t-tests,Spearman correlation analysis,and least absolute shrinkage and selection operator(LASSO)regression.Three distinct predictive models were established,including intratumoral,intratumoral plus peritumoral edema,and fusion models.Nomograms were employed to display the predictions of the 3-year survival period,while Kaplan-Meier(KM)plots were utilized to illustrate the survival outcomes across different groups.Results:Statistically significant differences were observed in the isocitrate dehydrogenase(IDH)mutation status and Rad-score between the recurrence and non-recurrence groups,with P-values of 0.04 and<0.001,respectively.The final analysis included fifteen imaging radiomics features.The three models in the training set displayed area under the curve(AUC)values of 0.905,0.925,and 0.923,while in the test set,the corresponding AUC values were 0.859,0.866,and 0.897.The fusion model outperformed the others.KM analysis demonstrated no significant differences in survival time among patient groups in both the training and test sets.Conclusions:MRI-based imaging radiomics demonstrates promising predictive capability for postoperative recurrence in glioblastoma patients,while also offering an initial assessment of postoperative survival time.

关 键 词：胶质瘤 影像组学 机器学习 生存期 磁共振成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R730.264[医药卫生—诊断学]