Multiparametric MRI Along with Machine Learning Predicts Prognosis and Treatment Response in Pediatric Low-Grade Glioma

Abstract

In this study, we present a comprehensive radiogenomic analysis of pediatric low-grade gliomas (pLGGs), combining treatment-naive multiparametric MRI and RNA sequencing. We identified three immunological clusters using XCell enrichment scores, highlighting an 'immune-hot' group correlating with poorer prognosis, suggesting potential benefits from immunotherapies. A radiomic signature predicting immunological profiles showed balanced accuracies of 81.5% and 84.4% across discovery and replication cohorts, respectively. Our clinicoradiomic model predicted progression-free survival with concordance indices of 0.71 and 0.77 in these cohorts, and the clinicoradiomic scores correlated with treatment response (p = 0.001). We also explored germline variants and transcriptomic pathways related to clinicoradiomic risk, identifying those involved in tumor growth and immune responses. This is the first radiogenomic analysis in pLGGs that enhances prognostication by prediction of immunological profiles, assessment of patients' risk of progression, prediction of treatment response to standard-of-care therapies, and early stratification of patients to identify potential candidates for novel therapies targeting specific pathways.