Unique Opportunities and Challenges of Artificial Intelligence in Pediatric Imaging
Abstract Artificial intelligence (AI) in pediatric imaging is evolving rapidly but lags behind and remains fundamentally distinct from adult radiology AI, due to various factors, such as developmental variability, differing disease epidemiology, unique imaging constraints, and limited data ecosystems. This narrative review examines the biological, technical, and structural factors that shape pediatric imaging AI, summarizes current and emerging applications, and highlights key barriers to safe clinical translation. Evidence indicates that the most immediate and impactful applications of pediatric AI lie in image acquisition, dose reduction, and workflow optimization, while interpretive and prognostic tools remain variably mature, with bone age assessment and chest radiograph interpretation representing the most established use cases. However, broader deployment is constrained by small, fragmented datasets, lack of standardized reference labels, limited generalizability, insufficient prospective validation, and gaps in regulatory and clinical integration frameworks. Furthermore, adult-trained models frequently underperform in children, reinforcing the need for pediatric-specific design and evaluation. The review emphasizes the importance of ethical governance, stakeholder trust, and multidisciplinary collaboration in guiding implementation. Future progress will depend on shared data infrastructures, federated learning approaches, rigorous age-stratified validation, and outcome-focused evaluation aligned with pediatric care priorities. Overall, pediatric imaging AI should be regarded as a distinct clinical, technical, and ethical domain requiring evidence-driven, child-centered development to ensure safe, effective, and equitable adoption in practice.