LGG-19. MULTI-MODALITY CHARACTERIZATION OF PEDIATRIC LOW-GRADE GLIOMA: THE ADDED VALUE OF DIFFERENT DATA MODALITIES

Characterization of pediatric low-grade glioma (pLGG) remains a significant challenge in the field of neuro-oncology, with a need for more effective precision diagnostics. Multi-modal analysis, which incorporates data from varied sources, has the potential to provide a comprehensive understanding of the underlying biology of pediatric brain tumors. Despite its potential, the utilization of true multi-modality clustering remains limited in pediatric brain tumor research. In this study, we aimed to address this gap by using a clustering model that incorporated genomics, radiomics, and clinical variables (age, sex, and tumor location) to group patients into distinct clusters. 103 patients with pLGGs were included. Mutations data was derived from whole genome sequencing obtained through the PedCBioportal. Radiomic data was obtained from MR imaging through the Children’s Brain Tumor Network and included features from pre- and post-contrast T1, T2, FLAIR, and ADC sequences. Categorical variables included sex (male vs female), genetic mutation status for 10 selected genes (BRAF, FGFR1, TSC1, TSC2, NF1, MYB, EGFR, ALK, IDH1, and RB1) that are known to play a role in the pathogenesis of pLGG (mutated vs non-mutated), and tumor location (9 categories). Continuous variables included age (days) and radiomic features. All numerical variables were normalized and reduced into principal components that captured 90% of the variance in the data prior to clustering. Various clustering iterations were performed incorporating combinations of radiomic, genomic, and clinical data. Our models identified two distinct clusters and the PFS differences between clusters approached statistical significance with the integration of information from all modalities when compared to any combination of subsets of data, highlighting the complementary value of these modalities in providing a comprehensive characterization of pLGG. This study provides preliminary evidence for the utility of multi-modality data clustering in improving our understanding of pLGG and supports further investigation into this approach.