Pediatric brain tumors are rare and still represent the most common solid tumors in children and the leading cause of cancer-related mortality in the pediatric population. Compared to adult brain tumors, they exhibit distinct biology, anatomy, and clinical behavior, posing unique diagnostic and therapeutic challenges. Artificial intelligence (AI) methods have the potential to improve diagnosis, disease monitoring, and treatment response assessment, but progress in pediatric neuro-oncology has been hampered by the lack of large, standardized, and publicly accessible datasets. We introduce the Brain Tumor Segmentation in Pediatrics (BraTS-PEDs) dataset, the first large-scale open-access benchmark data dedicated to pediatric brain tumor segmentation and analysis. The dataset comprises multiparametric MRI scans from 457 pediatric patients with high-grade gliomas, collected across multiple institutions and international consortia. Each case includes pre- and post-contrast T1-weighted, T2-weighted, and T2-FLAIR MRI sequences. Tumor subregions were annotated following the Response Assessment in Pediatric Neuro-Oncology (RAPNO) recommendations through a semi-automated process combining pediatric-specific auto-segmentation and expert manual refinement by neuroradiologists. The dataset is partitioned into training (n = 257), validation (n = 91), and hidden testing (n = 109) subsets to support reproducible benchmarking. BraTS-PEDs is the first large-scale, standardized resource for developing and evaluating AI algorithms in pediatric neuro-oncology. It provides a foundation for reproducible method comparison, model generalization across institutions, and future integration of imaging with molecular and clinical data for precision medicine applications.
Kazerooni et al. (Wed,) studied this question.