Multi-parametric MRI-based machine learning model for prediction of WHO grading in patients with meningiomas

作者：Zhen Zhao, Chuansheng Nie, Lei Zhao, Dongdong Xiao, Jianglin Zheng, Hao Zhang, Peng-Fei Yan, Xiaobing Jiang, Hongyang Zhao · 发表于：European Radiology · 年份：2023 · DOI：10.1007/s00330-023-10252-8 · 被引用次数：34 · 研究领域：Meningioma and schwannoma management、Brain Metastases and Treatment、Glioma Diagnosis and Treatment

OBJECTIVE: The purpose of this study was to develop and validate a nomogram combined multiparametric MRI and clinical indicators for identifying the WHO grade of meningioma. MATERIALS AND METHODS: Five hundred and sixty-eight patients were included in this study, who were diagnosed pathologically as having meningiomas. Firstly, radiomics features were extracted from CE-T1, T2, and 1-cm-thick tumor-to-brain interface (BTI) images. Then, difference analysis and the least absolute shrinkage and selection operator were orderly used to select the most representative features. Next, the support vector machine algorithm was conducted to predict the WHO grade of meningioma. Furthermore, a nomogram incorporated radiomics features and valuable clinical indicators was constructed by logistic regression. The performance of the nomogram was assessed by calibration and clinical effectiveness, as well as internal validation. RESULTS: Peritumoral edema volume and gender are independent risk factors for predicting meningioma grade. The multiparametric MRI features incorporating CE-T1, T2, and BTI features showed the higher performance for prediction of meningioma grade with a pooled AUC = 0.885 (95% CI, 0.821-0.946) and 0.860 (95% CI, 0.788-0.923) in the training and test groups, respectively. Then, a nomogram with a pooled AUC = 0.912 (95% CI, 0.876-0.961), combined radiomics score, peritumoral edema volume, and gender improved diagnostic performance compared to radiomics model or clinical m...