Background and purpose: Cancers show heterogeneity at various levels, from genome to radiologic imaging. This study aimed to explore the interplay between genomic, transcriptomic, and radiophenotypic data in pediatric low-grade glioma (pLGG), the most common group of brain tumors in children.
Materials and methods: We analyzed data from 201 patients with pLGG in the Children's Brain Tumor Network by using principal component analysis and K-means clustering on 881 radiomic features, along with clinical variables (age, sex, tumor location), to identify imaging clusters and examine their association with pLGG classifications as described in the World Health Organization Classification of Tumors of the Central Nervous System, 5th edition, 2021 (WHO 2021 CNS 5). To determine the transcriptome pathways linked to imaging clusters, we employed a supervised machine-learning model with elastic net logistic regression based on the pathways identified through gene set enrichment and gene coexpression network analyses.
Results: Three imaging clusters with distinct radiomic characteristics were identified. BRAF V600E mutations were primarily found in imaging cluster 3, while KIAA1549::BRAF fusion occurred in subtype 1. The model's predictive accuracy was 0.77 for subtype 1, 0.78 for subtype 2, and 0.70 for subtype 3. Each imaging cluster exhibited unique molecular mechanisms: subtype 1 was linked to oxidative phosphorylation, PDGFRB, and interleukin signaling, whereas subtype 3 was associated with histone acetylation and DNA methylation pathways, related to BRAF V600E pLGGs.
Conclusions: Our radiogenomics study indicates that the intrinsic molecular characteristics of tumors correlate with distinct imaging subgroups in pLGG, paving the way for future multimodal investigations that may enhance understanding of disease progression and targetability.
© 2025 by American Journal of Neuroradiology.