Noninvasive Prediction of Histological Grading in Pediatric Low-Grade Gliomas Using Preoperative T2-FLAIR Radiomics Features

To investigate the clinical application value of radiomics based on magnetic resonance T2-fluid attenuated inversion recovery (FLAIR) sequence images to distinguish pediatric low-grade gliomas of histological grades 1 and 2. A retrospective study of pediatric low-grade gliomas treated in our institution from April 2017 to July 2021. The histological grading follows the 2021 WHO (World Health Organization) classification of tumors of the central nervous system and contains the necessary molecular phenotype information. The 3D slicer (https://slicer.org/) is used to outline volume of interest based on T2-FLAIR sequence and extract three-dimensional imaging features. All enrolled cases are randomly assigned to training set and test set according to 7:3; SMOTE (Synthetic Minority Oversampling Technique) method was used to balance the data of the training set, and then min-max normalization was used to normalize the data of the radiomics features. Dimension reduction and screening were carried out through Pearson correlation coefficients, analysis of variance (ANOVA), and least absolute shrinkage and selection operator (LASSO) algorithms for the radiomics features. The best binary logistic regression model is established by using the best subset regression, and the receiver operating characteristic curve, calibration curve and decision curve are used to analyze and evaluate the model. A total of 113 patients were enrolled, 79 in the training set and 34 in the test set. There was no significant difference in sex and age between WHO grade 1 and 2 pediatric low-grade gliomas. A total of 1643 radiomics features were extracted from T2-FLAIR images, and finally 9 features were selected to construct a binary logistic regression model. The areas under the curve were 0.902 (95% confidence interval, 0.814–0.967) and 0.831 (95% confidence interval, 0.613–0.975) for the training and test sets, with sensitivities of 86.70% and 85.7% and specificities of 81.3% and 59.3%, respectively. For model calibration, the mean absolute errors were 0.054 and 0.058 for the training and test sets, respectively. The decision curve analysis showed clinical gains for using the model in both the training and testing sets. The T2-FLAIR radiomics model can be used for preoperative identification of grade 1 and grade 2 pediatric low-grade gliomas.