Feasibility of T2-Weighted MRI Radiomics for Initial Risk Stratification in Pediatric Neuroblastoma

Purpose: The purpose of this study was to evaluate the feasibility of magnetic resonance imaging (MRI)-based radiomics derived from routine T2-weighted imaging for initial risk stratification in pediatric neuroblastoma (NB) and to explore its potential role as a noninvasive adjunct to established clinical and molecular classification systems. Methods: In this retrospective, single-center pilot study, 45 children with newly diagnosed NB (2015–2024) were analyzed. Primary tumors were manually segmented on baseline axial T2-weighted MRI. A total of 107 Image Biomarker Standardization Initiative (IBSI)-compliant radiomic features were extracted. Supervised machine learning classifiers (Random Forest, XGBoost) and dimensionality reduction approaches (principal component analysis PCA, linear discriminant analysis LDA) combined with K-means clustering were evaluated. Model performance was assessed using stratified cross-validation and an independent test set. Reporting adhered to the Checklist for Evaluation of Radiomics Research (CLEAR). Results: Fifteen patients (33%) were classified as high-risk (HR) and 30 (67%) as non-high-risk (NHR) according to International Neuroblastoma Risk Group (INRG) criteria. The highest classification performance was achieved using LDA followed by K-means clustering, with a test accuracy of 77.8%, sensitivity of 64.7%, and specificity of 85.7%. Radiomic classification agreed with conventional risk stratification in 77.8% of cases. The analysis relied exclusively on T2-weighted imaging, supporting workflow feasibility without requiring contrast administration or advanced MRI sequences. Conclusions: In this single-center pilot study, T2-weighted MRI radiomics demonstrated feasibility for noninvasive initial risk stratification in pediatric NB. Although limited by sample size and the lack of external validation, these findings support further multicenter investigations of radiomics as an adjunctive imaging biomarker during early diagnostic evaluation.