RESEARCH presented at the World Congress on Osteoporosis (WCO) suggests that artificial intelligence could transform how low bone mineral density is detected—using something as routine as a chest X-ray.
The multi-centre study evaluated a novel AI system, OsteoPredict, designed to identify patients with osteoporosis-level bone loss (T-score ≤ -2.5) without the need for dedicated bone scans.
Turning Routine Imaging into Screening Tools
Dual-energy X-ray absorptiometry (DXA) remains the gold standard for diagnosing low bone mineral density, but access and uptake remain inconsistent globally. The researchers aimed to address this gap by leveraging chest X-rays—one of the most commonly performed imaging tests in clinical practice.
Using a federated learning approach, the AI model was trained on nearly 6,000 images while preserving patient data privacy across institutions. It was then externally validated on over 3,000 patients from two independent medical centres.
Strong Accuracy Across Populations
The results were hard to ignore. The system achieved an area under the curve (AUROC) of 0.942, indicating excellent diagnostic performance. Sensitivity reached 91.0%, meaning the model was highly effective at identifying individuals with low bone mineral density, while specificity stood at 82.7%.
Importantly, performance remained consistent across both validation sites and between genders, suggesting strong generalisability.
High-Risk Groups Identified Effectively
One of the more clinically relevant findings was the model’s performance in older adults. In patients aged 60 years and above—those at highest risk of osteoporosis—the system maintained high sensitivity, minimising the chance of missed diagnoses.
This matters, because osteoporosis is often underdiagnosed until fractures occur, by which point the clinical and economic burden is significantly higher.
A Scalable Screening Opportunity
By using chest X-rays already performed for other indications, this AI approach opens the door to opportunistic screening at scale. Instead of relying solely on targeted DXA referrals, healthcare systems could flag at-risk patients automatically during routine imaging workflows.
What Comes Next
While the results are promising, the study was retrospective, and prospective validation will be needed before widespread implementation. Integration into clinical pathways will also require careful consideration, particularly around follow-up testing and cost-effectiveness.
Still, the direction is clear. If this kind of technology holds up in real-world settings, it could quietly shift osteoporosis detection from reactive to proactive, without adding extra burden to patients or clinicians.
