# Artificial Intelligence in Paediatric and Adolescent Fracture Detection: A Systematic Review and Meta-Analysis

## TL;DR

AI can detect fractures in children more accurately than human interpretation and improves clinician performance when used as an aid.

## Contribution

This study is the first to systematically evaluate AI's diagnostic performance for pediatric fractures and its impact on clinician accuracy.

## Key findings

- Standalone AI showed higher sensitivity than human interpretation for detecting pediatric fractures.
- AI-assisted diagnosis improved clinicians' sensitivity in fracture detection.
- AI is a promising tool for enhancing diagnostic accuracy and efficiency in pediatric fracture detection.

## Abstract

Fractures are among the most common injuries in children, yet their radiographic detection is challenging due to the unique anatomy of the developing skeleton, leading to significant diagnostic errors. To address this, a systematic review and meta-analysis was conducted to evaluate how accurately and efficiently artificial intelligence (AI) detects fractures in children and adolescents. Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines, a systematic search of PubMed, EMBASE, and Web of Science identified 11 studies published between 2019 and 2024 evaluating AI for detecting appendicular skeletal fractures in patients under 21 years. A meta-analysis revealed that standalone AI demonstrated a statistically significantly higher sensitivity compared to human interpretation (mean difference: 0.04, 95% CI [0.02, 0.07], p = 0.0005) with non-inferior specificity. Furthermore, AI-assisted diagnosis led to a statistically significant improvement in clinician sensitivity (mean difference: 0.07, p = 0.003). To sum up, AI exhibits high diagnostic performance for paediatric fractures and serves as a promising adjunct tool to enhance clinical efficiency and accuracy; however, further large-scale, multi-centre prospective trials are required to validate its real-world applicability and address current limitations before widespread adoption.