AI-enhanced routine CT for opportunistic osteoporosis screening and risk assessment.

Abstract

Osteoporosis-related fractures create a large and often preventable burden, yet many individuals undergoing CT for unrelated care do not receive skeletal assessment. This review evaluates artificial intelligence (AI)-enabled use of routine CT for opportunistic case-finding and risk assessment, with emphasis on the boundary between promising retrospective performance and safe clinical implementation. Automated vertebral segmentation and CT-derived bone-density estimation can reduce manual workload, and published diagnostic studies report high discrimination in selected cohorts. However, pooled estimates are heterogeneous because the reference standard, anatomical target, CT acquisition, reconstruction, contrast status, calibration method, prevalence, and validation design differ substantially across studies. These differences preclude a universal CT threshold or a universal minimum detector-row requirement; detector-row configuration should be considered together with the broader acquisition and reconstruction parameters. For implementation, AI should be deployed as a validated triage tool-not as a stand-alone diagnostic replacement for clinical assessment and indicated DXA/QCT. A credible pathway requires a defined CT-protocol eligibility check; pre-specified vertebral region of interest and exclusions; protocol- and contrast-aware quantification; scanner-specific quality assurance and calibration monitoring where applicable; external validation in representative populations; and an explicit route from an AI alert to confirmatory testing and treatment evaluation. AI may also enable concurrent vertebral-fracture and body-composition assessment, but clinical utility must be shown through downstream testing, treatment, fracture, workload, equity, and cost outcomes. The near-term priority is therefore not higher retrospective diagnostic performance alone, but reproducible, transparent, and prospectively evaluated integration into radiology workflows.