Combined Evaluation of T1 and Diffusion MRI Improves the Noninvasive Prediction of H3K27M Mutation in Brainstem Gliomas

Objective: To establish an individualized predictive model to identify patients with brainstem gliomas (BSGs) at high risk of H3K27M mutation with the inclusion of diffusion MRI (dMRI). Methods: A cohort of 126 patients with BSGs were respectively included. Local tumor radiomics features were extracted from T1 images, and global brain connectomics features were extracted from dMRI data. A machine learning-based individualized H3K27M mutation prediction model was generated with a nested cross validation strategy. Additionally, two predictive signatures were established using the elastic net method, and simplified logistic models were built using multivariable logistic regression analysis. Results: 21 radiomics features and 52 topological properties of brain structural connectivity network were selected to construct a machine learning-based H3K27M mutation prediction model, which achieved an accuracy of 92.11% (AUC = 0.9246) in the validation cohort. T1- and dMRI-based signatures were generated and the combined multivariate logistic model was built, which achieved an AUC of 0.8783 in the validation cohort. Conclusion: dMRI is valuable in prediction of H3K27M mutation in BSGs. Combining multiple MRI sequences and clinical features, the established models have good performance.