AI-Driven opportunistic osteoporosis screening: the skeletal failure paradigm for trustworthy vertebral fracture triage

Opportunistic CT screening for osteoporosis is moving toward clinical implementation, but progress is constrained by the absence of a standardized, machine learning compatible ground truth for vertebral fractures. Traditional semi-quantitative grading systems show only moderate inter-observer agreement, with ambiguity concentrated at mild deformities, introducing label noise that degrades AI triage performance. This invited review synthesizes the current evidence and proposes a hypothesis-generating framework; the Skeletal Failure paradigm is a proposal requiring prospective validation, not an established standard. Against this background, we outline the current landscape of AI-driven opportunistic CT tools and identify label inconsistency as a central barrier to reliable deployment. Drawing on this synthesis and the author’s clinical and research experience, we propose a binary, action-oriented classification—“Skeletal Failure”—defined by unequivocal endplate disruption or >20% vertebral height loss, providing a potentially reproducible structural endpoint. When combined with phantomless CT trabecular Hounsfield unit measurements, this framework is proposed as a candidate ML-ready substrate for automated triage and structured reporting. The Skeletal Failure paradigm is designed to support efficient human-in-the-loop triage, reduce radiologist cognitive load, and facilitate population-level fracture prevention. It may be particularly useful in resource-constrained settings, including the Middle East and North Africa, where DXA availability is limited and WHO T-score thresholds underperform in non-Caucasian populations. By converting incidental CT findings into consistent, machine-readable risk profiles, this approach aims to better align algorithmic detection with clinically actionable pathways, especially in older adults at risk of vertebral fragility and underdiagnosis.