Diagnostic accuracy of deep learning models for detecting wrist and distal forearm fractures in pediatric patients: A systematic review and meta-analysis.

PURPOSE: Pediatric wrist and distal forearm fractures are common but can be challenging to diagnose due to age-related skeletal differences and complex fracture patterns. Deep learning models may improve fracture detection; however, their diagnostic performance in pediatric wrist injuries remains unclear. This study aimed to evaluate the diagnostic accuracy and clinical utility of deep learning models for detecting wrist and distal forearm fractures in children. METHODS: In this systematic review and diagnostic meta-analysis, MEDLINE, Web of Science, and Scopus were searched from inception to July 2026. Quality appraisal was performed using the QUADAS-2 tool. Diagnostic parameters were estimated using Meta-Disc software and the MIDAS package in Stata. Meta-regression analyses were performed to explore potential sources of heterogeneity. RESULTS: Deep learning models achieved a pooled sensitivity of 0.93 (95% CI, 0.92-0.95) and specificity of 0.85 (95% CI, 0.83-0.87) for fracture detection. The positive likelihood ratio (PLR) and negative likelihood ratio (NLR) were 6.83 (95% CI, 5.27-8.84) and 0.08 (95% CI, 0.05-0.12), respectively. The area under the summary receiver operating characteristic (SROC) curve was 0.96. CONCLUSION: Deep learning models demonstrate high diagnostic performance for detecting wrist and distal forearm fractures in pediatric patients, with particular potential as decision-support tools for fracture exclusion. Diagnostic performance varied by imaging modality and model type, with ultrasonography-based models showing higher specificity and commercially available models showing higher sensitivity. However, limited evidence, methodological heterogeneity, and insufficient external validation highlight the need for prospective multicenter studies before widespread clinical implementation.