机器学习结合影像组学特征鉴别间变性胶质细胞瘤和胶质母细胞瘤  

A model combined machine learning with imaging omics characteristics in differentiating anaplastic glioma from glioblastoma

作　　者：王策 钱增辉[1] 蔡泽豪 康庄 陈宝师[1] Wang Ce;Qian Zenghui;Cai Zehao;Kang Zhuang;Chen Baoshi(Department of Neurosurgery,Tiantan Hospital of Beijing,Capital Medical University,Beijing 100071,China)

机构地区：[1]首都医科大学附属北京天坛医院神经外科,北京100071

出　　处：《中华神经医学杂志》2020年第3期224-228,共5页Chinese Journal of Neuromedicine

基　　金：北京天坛医院青年基金(2018-YQN-8)。

摘　　要：目的结合机器学习与影像组学特征构建预测间变性胶质细胞瘤和胶质母细胞瘤的模型并进行验证。方法回顾性收集首都医科大学附属北京天坛医院神经外科自2005年8月至2012年8月收治的241例经病理证实的间变性胶质细胞瘤或胶质母细胞瘤患者的影像学资料。按随机数字表法分为训练组(n=140)和验证组(n=101)。使用MRIcron软件在训练组患者术前T1增强影像上勾勒肿瘤边界,运用Matlab软件提取肿瘤区域的影像组学特征,应用最低绝对收缩和选择算子(LASSO)回归模型筛选最佳影像组学特征,基于所选特征通过支持向量机(SVM)分类器建立胶质瘤鉴别模型,使用受试者工作特征(ROC)曲线评价模型对间变性胶质细胞瘤或胶质母细胞瘤的预测效果。结果241例胶质瘤患者中,经病理证实为间变性胶质细胞瘤101例、胶质母细胞瘤140例。训练组和验证组中,胶质母细胞瘤、间变性胶质细胞瘤患者的年龄差异均有统计学意义(P<0.05);胶质母细胞瘤和间变性胶质细胞瘤患者的性别分布,肿瘤位置,肿瘤坏死、水肿所占比例差异均无统计学意义(P>0.05)。Matlab软件共提取431个影像组学特征,LASSO回归模型筛选出11个最佳影像组学特征。ROC曲线分析显示SVM分类器建立的模型预测训练组患者胶质瘤类型的曲线下面积(AUC)为0.942,预测验证组患者胶质瘤类型的AUC为0.875。结论结合机器学习与影像组学特征构建的模型可有效鉴别间变性胶质细胞瘤和胶质母细胞瘤。Objective To construct and validate a prediction model combined machine learning with imaging omics characteristics in differentiating anaplastic glioma from glioblastoma.Methods Imaging data of 241 patients with anaplastic glioma or glioblastoma,confirmed by pathology in our hospital from August 2005 to August 2012,were retrospectively collected.These patients were divided into a training group(n=140)and a verification group(n=101)according to random number table method.MRIcron software was used to delineate tumor boundaries of patients from the training group on preoperative T1 enhanced MR imaging.The regions of interest(ROIs)were outlined on preoperative T1 enhanced MR imaging,and the radiomic features were extracted from ROIs by Matlab software.Least absolute shrinkage and selection operator(LASSO)regression model was used to screen the features,and then,the selected features were used to construct the prediction model by support vector machine(SVM)classifier.The area under the curve(AUC)of receiver operating characteristic(ROC)curve was used to evaluate the predictive efficacy of the model.Results In these 241 patients,101 were with anaplastic glioma and 140 were with glioblastoma confirmed by pathology.In the training group and validation group,there was statistical difference in age between patients with anaplastic glioma and glioblastoma(P<0.05);there was no significant difference in gender distribution,tumor location,and percentages of tumor necrosis or edema between patients with anaplastic glioma and glioblastoma(P>0.05).Totally,431 radiomic features were extracted;11 radiomic features were screened by LASSO regression model and the prediction model was established.The AUC of ROC curve was 0.942 and 0.875,respectively,in the training group and validation group.Conclusion The prediction model combined machine learning and imaging omics characteristics can effectively discriminate anaplastic glioma from glioblastoma.

关 键 词：机器学习 影像组学 MRI增强成像 间变性胶质细胞瘤 胶质母细胞瘤 

分 类 号：R73[医药卫生—肿瘤]