Diagnostic accuracy of AI-assisted versus independent physician interpretation for bone fractures: a systematic review and meta-analysis

Abstract

We systematically evaluated the diagnostic performance of artificial intelligence (AI)-assisted interpretation versus independent physician assessment for fracture detection. Adhering to PRISMA-DTA guidelines, we searched PubMed and Web of Science for original studies published up to September 17, 2025. Quality was assessed utilizing the QUADAS-3 framework. A bivariate random-effects model pooled diagnostic metrics. Accuracy was assessed by summary receiver operating characteristic (SROC) curves and its area under the curve (AUC). Mixed-effects meta-regression explored heterogeneity. Of 450 records retrieved, after excluding 118 duplicates, the remaining 332 records were screened by title/abstract, identifying 38 candidates. Following full-text evaluation, 28 studies met the inclusion criteria, 17 of them suitable for meta-analysis. Compared to unassisted diagnosis, AI significantly improved pooled sensitivity (87%, 95% confidence interval [CI]: 84–89%, versus 73%, 95% CI: 69–78%) and maintained high specificity (95%, 95% CI: 92–97%, versus 94%, 95% CI: 89–96%). The SROC-AUC increased from 0.849 to 0.929. Subgroup analysis revealed junior clinicians derived the greatest benefit, exhibiting a 21% absolute sensitivity increase. Multivariate meta-regression, explaining 52.2% of heterogeneity, identified two-dimensional x-ray and junior physician status (p ≤ 0.005) as independent predictors of a lower absolute AI-assisted diagnostic ceiling. AI assistance is an effective diagnostic adjunct, substantially improving sensitivity and mitigating the experience gap for junior clinicians. However, multivariate evidence confirms AI cannot fully supersede the absolute diagnostic ceiling dictated by foundational clinical expertise and two-dimensional radiography’s physical limitations. Future workflows must optimize human-AI collaboration while maintaining a low threshold for cross-sectional imaging. Question Accurate fracture interpretation remains a significant challenge for non-specialists. This study quantifies how human-machine collaboration effectively mitigates clinical experience gaps and improves radiologic diagnosis. Findings AI assistance significantly increased overall pooled diagnostic sensitivity from 73% to 87% without compromising specificity, providing the greatest absolute diagnostic benefit to junior clinicians. Relevance Statement Integrating AI into high-pressure workflows substantially reduces missed fractures and safely bridges the experience gap for junior clinicians, ultimately optimizing patient triage and operational efficiency.