本研究旨在开发一种基于磁共振（MR）放射组学特征的集成学习（EL）方法，以在术前区分颅内室外室管膜瘤（IEE）和胶质母细胞瘤（GBM）。这项回顾性研究纳入了2016年6月至2021年6月期间经组织病理学确诊的IEE和GBM患者。从T1加权成像（T1WI）和T2加权成像（T2WI）序列图像中提取放射组学特征，并使用EL方法和分类模型构建分类模型。逻辑回归（LR）。使用受试者工作特征（ROC）曲线、校准曲线和决策曲线分析来评估模型的效率。基于 T1WI 和 T2WI 图像的临床参数和放射组学特征的组合 EL 模型表现出良好的辨别能力，受试者工作特征曲线下面积 (AUC) 为 0.96 (95% CI 0.94-0.98)，特异性为 0.84 ，训练集中的准确度为 0.92，灵敏度为 0.95，验证集中的 AUC 为 0.89（95% CI 0.83-0.94），特异性为 0.83，准确度为 0.81，灵敏度为 0.74。 EL模型的判别效能显着高于LR模型。观察到 EL 模型具有良好的校准性能和临床适用性。 EL 模型结合了术前基于 MR 的肿瘤放射组学和临床数据，在术前区分 IEE 和 GBM 方面表现出很高的准确性和敏感性，这可能有助于这些脑肿瘤的临床管理。© 2024 John Wiley

This study aims to develop an ensemble learning (EL) method based on magnetic resonance (MR) radiomic features to preoperatively differentiate intracranial extraventricular ependymoma (IEE) from glioblastoma (GBM). This retrospective study enrolled patients with histopathologically confirmed IEE and GBM from June 2016 to June 2021. Radiomics features were extracted from T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI) sequence images, and classification models were constructed using EL methods and logistic regression (LR). The efficiency of the models was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis. The combined EL model, based on clinical parameters and radiomic features from T1WI and T2WI images, demonstrated good discriminative ability, achieving an area under the receiver operating characteristics curve (AUC) of 0.96 (95% CI 0.94-0.98), a specificity of 0.84, an accuracy of 0.92, and a sensitivity of 0.95 in the training set, and an AUC of 0.89 (95% CI 0.83-0.94), a specificity of 0.83, an accuracy of 0.81, and a sensitivity of 0.74 in the validation set. The discriminative efficacy of the EL model was significantly higher than that of the LR model. Favorable calibration performance and clinical applicability for the EL model were observed. The EL model combining preoperative MR-based tumor radiomics and clinical data showed high accuracy and sensitivity in differentiating IEE from GBM preoperatively, which may potentially assist in clinical management of these brain tumors.© 2024 John Wiley & Sons Ltd.