Radiomic Features of the Edema Region May Contribute to Grading Meningiomas With Peritumoral Edema

Abstract

Background: Meningiomas are frequently accompanied by peritumoral edema (PTE). The potential value of radiomic features of edema region in meningioma grading has not been investigated.

Purpose: To investigate whether radiomic features of edema region contribute to grading meningiomas with PTE.

Study type: Retrospective.

Population: A total of 444 patients including 196 grade II and 248 WHO grade I meningiomas: 356 patients for training, 88 for validation.

Field strength/sequence: A 1.5-T/3.0-T, noncontrast T1-weighted (T1WI), T2-weighted (T2WI), contrast-enhanced T1-weighted (T1CE) spin echo sequences.

Assessment: A total of 851 radiomic features were extracted from each sequence on each region (tumor and edema region). These features were integrated by region respectively. Three subsets of clinical-radiomic features were constructed by joining clinical information (sex, age, tumor volume, and edema volume) and radiomic features of three regions: tumor, edema, and combined subsets. For each subset, features were filtered by the least absolute shrinkage and selection operator (LASSO) and Random Forest algorithm. Top 20 features of each subset were finally selected.

Statistical tests: Stochastic Gradient Boosting, Random Forest, and Bagged AdaBoost predictive models were built based on each subset. Discriminative abilities of models were quantified using receiver operating characteristics (ROC) and the area under the curve (AUC). A P value < 0.05 was considered statistically significant.

Results: Random Forest model based on combined subset (AUC [95% CI] = 0.880 [0.807-0.953]) had the best discriminative ability in grading meningiomas among the final models. The best model of edema subset and tumor subset were Random Forest model (AUC [95% CI] = 0.864 [0.791-0.938]) and Stochastic Gradient Boosting model (AUC [95% CI] = 0.844 [0.760-0.928]), respectively.

Data conclusion: Radiomic features of edema region may contribute to grading meningiomas with PTE. The Random Forest model based on combined subset surpasses the best model based on tumor or edema subset regarding grading meningiomas with PTE.

Evidence level: 4 TECHNICAL EFFICACY: Stage 3.

Keywords: edema; meningioma; neoplasm grading; radiomics.