| 刘书涵,李锦龙,周青.对比增强液体衰减反转恢复序列MRI影像组学评估成人弥漫性低级别胶质瘤1p/19q状态[J].中国医学影像技术,2022,38(10):1470~1475 |
| 对比增强液体衰减反转恢复序列MRI影像组学评估成人弥漫性低级别胶质瘤1p/19q状态 |
| Radiomics of contrast enhancement-fluid attenuation inversion recovery sequence MRI for evaluation on 1p/19q status of adult diffuse low grade glioma |
| 投稿时间：2022-05-05 修订日期：2022-07-03 |
| DOI：10.13929/j.issn.1003-3289.2022.10.006 |
| 中文关键词: 胶质瘤 基因缺失 磁共振成像 影像组学 机器学习 |
| 英文关键词:glioma gene deletion magnetic resonance imaging radiomics machine learning |
| 基金项目: |
|
| 摘要点击次数: 2099 |
| 全文下载次数: 477 |
| 中文摘要: |
| 目的 观察对比增强液体衰减反转恢复（CE-FLAIR）序列MRI影像组学模型判断成人弥漫性低级别胶质瘤（DLGG）1p/19q状态的价值。方法 纳入135例成人DLGG患者、含81例1p/19q共缺失，经分层抽样按7 ∶ 3比例将其分为训练集（n=95）及验证集（n=40）。基于训练集CE-FLAIR数据提取、筛选DLGG 1p/19q共缺失影像组学特征，构建支持向量机（SVM）、随机森林（RF）、极限梯度提升（XGBoost）、轻量梯度提升机（LightGBM）及逻辑回归（LR）模型；绘制受试者工作特征曲线，计算曲线下面积（AUC），评价影像组学模型判断训练集及验证集DLGG 1p/19q状态的价值，并以DeLong检验进行比较。结果 共提取851个影像组学特征，以Mann-Whitney U检验筛选出74个差异有统计学意义者，再经5折交叉验证的最小绝对收缩和选择算子选出12个与1p/19q状态显著相关者；以之构建的SVM、RF、XGBoost、LightGBM及LR模型评价训练集DLGG 1p/19q状态的AUC分别为0.89、0.97、0.97、0.96及0.85，评估验证集的AUC分别为0.86、0.92、0.93、0.92及0.78。验证集中，LR模型的AUC低于SVM、RF、XGBoost及LightGBM （Z=2.981、3.136、3.014、2.827，P均<0.05），而后四者间AUC差异均无统计学意义（P均>0.05）；RF模型准确率最高，为88.24%。结论 基于CE-FLAIR影像组学模型可有效评估成人DLGG1p/19q状态；SVM、RF、XGBoost及LightGBM模型效能均较高，以RF模型准确率最高。 |
| 英文摘要: |
| Objective To observe the value of radiomics models of contrast enhancement-fluid attenuation inversion recovery (CE-FLAIR) sequence MRI for judging 1p/19q status in adult diffuse low grade gliomas (DLGG). Methods Totally 135 adult DLGG patients were enrolled, including 81 cases with 1p/19q co-deletion. The patients were divided into training set (n=95) or validation set (n=40) at the ratio of 7:3. Based on data of CE-FLAIR in the training set, radiomics features of DLGG 1p/19q co-deletion were screened and extracted, and the support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and logistic regression (LR) models were constructed. The receiver operating characteristic curve was drawn, the area under the curve (AUC) was calculated to evaluate the value of radiomics models for judging DLGG 1p/19q status, and the DeLong test was used to compare AUC of the models. Results A total of 851 features were extracted, 74 differential features were screened with Mann-Whitney U test, and 12 features significantly related to 1p/19q status were screened with 5-fold cross-validation of least absolute shrinkage and selection operator. AUC of SVM, RF, XGBoost, LightGBM and LR models for evaluating DLGG 1p/19q status of training set was 0.89, 0.97, 0.97, 0.96 and 0.85, respectively, of validation set was 0.86, 0.92, 0.93, 0.92 and 0.78, respectively. In validation set, AUC of LR model was the lower than that of SVM, RF, XGBoost and LightGBM model (Z=2.981, 3.136, 3.014, 2.827, all P<0.05), but no significant difference was found among SVM, RF, XGBoost and LightGBM model (all P>0.05). RF model had the highest accuracy (88.24%). Conclusion Radiomics models of CE-FLAIR could effectively evaluate 1p/19q status of adult DLGG. SVM, RF, XGBoost and LightGBM models had high efficacy, among which RF model had the highest accuracy. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
