Convolutional neural networks in paediatric fracture detection: pooled evidence from a systematic review and meta-analysis
-
By
-
March 21, 2026
Objective:
To synthesize the diagnostic accuracy of specific AI algorithms for paediatric appendicular fracture detection and assess limitations related to external validation and generalisability.
Key Findings:
- AI algorithms show promising diagnostic accuracy for detecting paediatric fractures, with some achieving over 90% sensitivity and specificity, which could significantly reduce missed diagnoses in clinical settings.
- Most existing research focuses on adult populations, highlighting a gap in paediatric applications that needs to be addressed.
- Limited external validation and generalisability of AI algorithms for paediatric fractures were noted, indicating a need for further research.
Interpretation:
AI has the potential to enhance fracture detection in children, addressing diagnostic challenges in busy emergency settings, but further research is needed to validate these algorithms in diverse clinical environments.
Limitations:
- Heterogeneity among studies and limited external validation, with specific examples of studies showing varying methodologies.
- Uncertainty regarding the generalisability of findings to broader paediatric populations.
- Most studies focused on specific anatomical regions, limiting comprehensive assessment and applicability.
Conclusion:
AI algorithms, particularly CNNs, may significantly improve the detection of fractures in children, but further studies are essential to establish their reliability and applicability in clinical practice, potentially transforming paediatric fracture management.
