From underdiagnosis to scalable screening: Applications of artificial intelligence for vertebral compression fracture detection - a systematic review and meta-analysis of diagnostic performance.
Authors
Affiliations (4)
Affiliations (4)
- Orthopedic Surgery Artificial Intelligence Laboratory (OSAIL), Department of Orthopedics, Mayo Clinic, Rochester, MN, USA.
- Department of Orthopedics, College of Medicine, National Taiwan University, Taipei, Taiwan.
- Mayo Clinic Libraries, Mayo Clinic, Rochester, MN, USA.
- Muscloskeletal Lab, Imperial College, London, UK.
Abstract
Vertebral compression fractures (VCFs) are common and often missed on routine imaging. We systematically reviewed the diagnostic performance of artificial intelligence (AI) for VCF detection. A systematic review and meta-analysis were conducted following PRISMA 2020 and PROSPERO registration by searching Ovid Medline, Embase, and Central from January 1, 2020, to November 18, 2025, for studies using AI models for VCF detection and diagnosis. Risk of bias was assessed with QUADAS-2. Studies with complete 2 × 2 data were pooled using random-effects meta-analysis, with subgroup analyses by modality. Fifty-one studies were included. Thirty-seven contributed to meta-analysis with around 259 190 participants and 45,346 VCF patients. Original radiology reports missed > 50% of incidental VCFs in many cohorts (up to 81%). The overall sensitivity of AI-assisted VCF detection was 85.0% (95% CI 84.2%-85.7%) and specificity 94.6% (95% CI 92.8%-95.9%). The positive and negative likelihood ratios were 15.4 and 0.16. At 22% pre-test probability, post-test probability was 77.7% after a positive AI result and 4.1% after a negative result. CT-based AI models performed best (sensitivity: 88.0%, specificity: 97.5%), while X-ray-based AI models showed lower sensitivity (79.2%) with high specificity (95.1%). AI models demonstrate high diagnostic accuracy for VCF detection, particularly in CT imaging. However, improved detection alone does not ensure better outcomes, as substantial treatment gaps persist. The clinical impact of AI depends on rigorous prospective external validation and integration into structured osteoporosis care pathways, linking automated identification to timely evaluation and secondary fracture prevention.