MRI-based machine learning models predict the malignant biological behavior of meningioma

Abstract

Abstract Background The WHO grade and Ki-67 index are independent indices to evaluate the malignant biological behavior of meningioma. This study aims to develop MRI-based machine learning models to predict the malignant biological behavior of meningioma from the perspective of the WHO grade, Ki-67 index, and their combination. Methods This multicenter, retrospective study included 216 meningioma patients (60 male and 156 female), of which 41 were classified as high-grade (WHO II/III) and 175 as low-grade (WHO I). The Ki-67 expression was classified into low-expressed (n=152) and high-expressed (n=64) groups with a threshold of 5%. Among them, there were 83 patients with malignant biological behavior whose WHO grade or Ki-67 index increased either or both. All patients were randomly divided into the training and test sets in a ratio of 7:3. Radiomic features were extracted from the maximum cross-sectional area (2D-ROI) and the whole tumor volume (3D-ROI) of the T1, T2-weighted, and T1 contrast-enhanced sequences, followed by five independent feature selections and eight classifiers. 240 prediction models were constructed for predict WHO grade, Ki-67 and their combination respectively. Models were evaluated by cross-validation in training set (n =151), suitable models were selected by compare the cross-validation area under the curves (AUC) and their relative standard deviations (RSD). Clinical and radiological features were collected and analyzed, and meaningful features were combined with radiomic features to establish the clinical-radiological-radiomic (CRR) models. Receiver operating characteristic (ROC) analysis were used to evaluate those models. Radiomic models and CRR models were compared by Delong test. Results 1218 and 1278 radiomic features were extracted from 2D-ROI and 3D-ROI of each sequence. The selected grade, Ki-67 and their combination radiomic models were T1CE-2D-LASSO-LR, T1CE-3D-LASSO-NB, and T1CE-2D-RFE-LR, with cross-validated AUCs on the training set were 0.878, 0.802, and 0.884, the RSDs were 0.055, 0.048, and 0.051, the test set AUCs were 0.807, 0.792, and 0.840, respectively. Heterogeneous enhancement was associated with high grade and Ki-67 status, while peritumoral edema was associated with high Ki-67 status. The Delong test shows that these significant radiological features did not significantly improve the predictive performance. The AUCs in the test set in predicting grade, Ki-67, and their combination were 0.811, 0.778, and 0.858, respectively. Conclusions This study demonstrated that MRI-based machine learning models could effectively predict the grade, Ki-67 index of meningioma. Models considering these two indices might be valuable for improving the predictive sensitivity and comprehensiveness of prediction of malignant biological behavior of meningiomas.