Department of Neurosurgery, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Department of Radiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Eur J Radiol. 2019 Dec;121:108714. doi: 10.1016/j.ejrad.2019.108714. Epub 2019 Oct 19.
The methylation status of the O6-methylguanine-DNA methyltransferase (MGMT) promoter has been proven to be a prognostic and predictive biomarker for lower grade glioma (LGG). This study aims to build a radiomics model to preoperatively predict the MGMT promoter methylation status in LGG.
122 pathology-confirmed LGG patients were retrospectively reviewed, with 87 local patients as the training dataset, and 35 from The Cancer Imaging Archive as independent validation. A total of 1702 radiomics features were extracted from three-dimensional contrast-enhanced T1 (3D-CE-T1)-weighted and T2-weighted MRI images, including 14 shape, 18 first order, 75 texture, and 744 wavelet features respectively. The radiomics features were selected with the least absolute shrinkage and selection operator algorithm, and prediction models were constructed with multiple classifiers. Models were evaluated using receiver operating characteristic (ROC).
Five radiomics prediction models, namely, 3D-CE-T1-weighted single radiomics model, T2-weighted single radiomics model, fusion radiomics model, linear combination radiomics model, and clinical integrated model, were built. The fusion radiomics model, which constructed from the concatenation of both series, displayed the best performance, with an accuracy of 0.849 and an area under the curve (AUC) of 0.970 (0.939-1.000) in the training dataset, and an accuracy of 0.886 and an AUC of 0.898 (0.786-1.000) in the validation dataset. Linear combination of single radiomics models and integration of clinical factors did not improve.
Conventional MRI radiomics models are reliable for predicting the MGMT promoter methylation status in LGG patients. The fusion of radiomics features from different series may increase the prediction performance.
O6-甲基鸟嘌呤-DNA 甲基转移酶（MGMT）启动子的甲基化状态已被证明是低级别胶质瘤（LGG）的预后和预测生物标志物。本研究旨在建立一种放射组学模型，以术前预测 LGG 中 MGMT 启动子的甲基化状态。
回顾性分析 122 例经病理证实的 LGG 患者，其中 87 例局部患者为训练数据集，35 例来自癌症成像档案的患者为独立验证数据集。从三维对比增强 T1（3D-CE-T1）加权和 T2 加权 MRI 图像中提取了 1702 个放射组学特征，包括形状特征 14 个、一阶特征 18 个、纹理特征 75 个和小波特征 744 个。使用最小绝对值收缩和选择算子算法选择放射组学特征，并使用多个分类器构建预测模型。使用受试者工作特征（ROC）评估模型。
构建了 5 种放射组学预测模型，即 3D-CE-T1 加权单放射组学模型、T2 加权单放射组学模型、融合放射组学模型、线性组合放射组学模型和临床综合模型。融合放射组学模型，由两个序列的串联构建而成，在训练数据集中表现最佳，准确率为 0.849，曲线下面积（AUC）为 0.970（0.939-1.000），在验证数据集中的准确率为 0.886，AUC 为 0.898（0.786-1.000）。单一放射组学模型的线性组合和临床因素的整合并没有提高。
常规 MRI 放射组学模型可可靠预测 LGG 患者 MGMT 启动子的甲基化状态。不同序列的放射组学特征融合可能会提高预测性能。
