Atlas-Based MRI Spatial Phenotyping in Pediatric Low-Grade Glioma

Pediatric low-grade gliomas (pLGGs) arise throughout the developing brain, and their anatomic distribution is related to age, molecular subtype, and extent of resection (EOR). In clinical studies, pLGG location is commonly represented by a single manually assigned anatomic category, which does not quantify the full spatial extent of tumors spanning multiple brain regions. We developed and evaluated a reproducible, publicly available pediatric atlas-based framework integrating voxel-wise probabilistic mapping with patient-level atlas-derived spatial features. In this retrospective study, preoperative MRI from 349 patients with pLGG was registered to the Haskins pediatric template using EasyReg. Probabilistic maps showed differences in tumor distribution by age, molecular subtype, and EOR, including greater optic pathway and suprasellar involvement at 0-2 years and greater cerebellar involvement in KIAA1549-BRAF fusion tumors. Atlas-derived spatial, tumor morphometric, and clinical features distinguished gross total resection/near total resection from subtotal resection or biopsy/no-surgery (Replication area under the curve, 0.864). Substituting 107 automatically computed regional overlap features for the manually assigned location variable in previously published clinico-imaging risk stratification models did not significantly alter discrimination (smallest P = 0.12). Together, these group- and patient-level analyses demonstrate the utility of the framework for quantitative spatial phenotyping in imaging and clinical studies of pLGG.