Ensemble learning‐based pretreatment MRI radiomic model for distinguishing intracranial extraventricular ependymoma from glioblastoma multiforme

Haoling He,Qianyan Long,Liyan Li,Yan Fu,Xueying Wang,Yuhong Qin,Muliang Jiang,Zeming Tan,Xiaoping Yi,Bihong T. Chen
DOI: https://doi.org/10.1002/nbm.5242
2024-08-23
NMR in Biomedicine
Abstract: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.
radiology, nuclear medicine & medical imaging,biophysics,spectroscopy
What problem does this paper attempt to address?