Neuroblastoma Risk Stratification

This study aimed to improve risk stratification and prognostic assessment for pediatric neuroblastoma by developing deep learning models that integrate multiphase enhanced CT images with clinical features. Using data from 202 patients, a Swin Transformer–based arterial phase model achieved the best risk stratification performance, while a combined Cox regression and randomized survival forest model showed strong prognostic accuracy with high AUC and C-index values. Overall, multimodal models outperformed clinical-only approaches, demonstrating their potential to support personalized treatment and precision medicine in neuroblastoma care.