Ruisong WANG, Xiaoman Ding, Wanyue Zhang, Tieliu Shi
Artificial intelligence (AI) possesses the transformative potential to reshape pediatric medicine, offering powerful tools for diagnosis, prognosis, and personalized therapy. This review focuses on three domains selected for their relative maturity in AI development and proximity to clinical translation-pediatric critical care, perinatal and neonatal medicine, and precision oncology-evaluating current evidence for clinical utility and outlining challenges to implementation. AI is demonstrating significant potential across these domains: in critical care, deep learning models outperform traditional scoring systems for dynamic prediction of adverse events; in perinatal and neonatal medicine, AI enhances prenatal ultrasonography and integrates multiomics data to guide complex therapies; and in oncology, radiomics, and genomic analysis enable non-invasive tumor characterization and personalized treatment strategies. However, significant hurdles remain. Foundational data challenges-including scarcity, heterogeneity, and limited sharing of pediatric data-are being addressed through transfer learning, federated learning, and synthetic data generation. Clinical translation is further impeded by algorithmic bias, the 'black box' problem, and the unique developmental physiology of children, which demands age-specific model validation. Future progress depends on multi-institutional collaboration, a research focus that extends beyond prediction to encompass causal inference and explainability, and the establishment of robust ethical, regulatory, and economic frameworks. Ultimately, responsible implementation of AI in pediatrics requires building systems that are not merely accurate but transparent, equitable, and trustworthy.