Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea.
Department of Clinical Epidemiology and Biostatistics, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea.
Neuro Oncol. 2021 Feb 25;23(2):324-333. doi: 10.1093/neuonc/noaa190.
Brain invasion by meningioma is a stand-alone criterion for tumor atypia in the 2016 World Health Organization classification, but no imaging parameter has yet been shown to be sufficient for predicting it. The aim of this study was to develop and validate an MRI-based radiomics model from the brain-to-tumor interface to predict brain invasion by meningioma.
Preoperative T2-weighted and contrast-enhanced T1-weighted imaging data were obtained from 454 patients (88 patients with brain invasion) between 2012 and 2017. Feature selection was performed from 3222 radiomics features obtained in the 1 cm thickness tumor-to-brain interface region using least absolute shrinkage and selection operator. Peritumoral edema volume, age, sex, and selected radiomics features were used to construct a random forest classifier-based diagnostic model. The performance was evaluated using the areas under the curves (AUCs) of the receiver operating characteristic in an independent cohort of 150 patients (29 patients with brain invasion) between 2018 and 2019.
Volume of peritumoral edema was an independent predictor of brain invasion (P < 0.001). The top 6 interface radiomics features plus the volume of peritumoral edema were selected for model construction. The combined model showed the highest performance for prediction of brain invasion in the training (AUC 0.97; 95% CI: 0.95-0.98) and validation sets (AUC 0.91; 95% CI: 0.84-0.98), and improved diagnostic performance over volume of peritumoral edema only (AUC 0.76; 95% CI: 0.66-0.86).
An imaging-based model combining interface radiomics and peritumoral edema can help to predict brain invasion by meningioma and improve the diagnostic performance of known clinical and imaging parameters.
在 2016 年世界卫生组织分类中，脑膜瘤侵犯脑是肿瘤非典型性的独立标准，但尚未有影像学参数足以预测其侵犯。本研究旨在开发和验证一种基于磁共振成像（MRI）的脑膜瘤脑侵犯的放射组学模型。
从 2012 年至 2017 年，共纳入 454 例患者（88 例存在脑膜瘤脑侵犯）的术前 T2 加权和对比增强 T1 加权成像数据。在 1cm 厚肿瘤-脑界面区域内提取 3222 个放射组学特征，采用最小绝对值收缩和选择算子（LASSO）进行特征选择。应用肿瘤周围水肿体积、年龄、性别和选定的放射组学特征构建随机森林分类器诊断模型。使用 2018 年至 2019 年期间的 150 例患者（29 例存在脑膜瘤脑侵犯）的独立队列，通过受试者工作特征曲线（ROC）下面积（AUC）评估模型的性能。
肿瘤周围水肿体积是脑膜瘤脑侵犯的独立预测因子（P < 0.001）。选择前 6 个界面放射组学特征和肿瘤周围水肿体积构建模型。该联合模型在训练集（AUC 0.97；95%CI：0.95-0.98）和验证集（AUC 0.91；95%CI：0.84-0.98）中对脑膜瘤脑侵犯的预测性能最高，并且优于仅基于肿瘤周围水肿体积的诊断性能（AUC 0.76；95%CI：0.66-0.86）。
一种结合界面放射组学和肿瘤周围水肿的基于影像学的模型有助于预测脑膜瘤脑侵犯，并提高已知临床和影像学参数的诊断性能。
