Research presented at the WCO-IOF-ESCEO Congress 2026 indicates that artificial intelligence (AI) may transform how clinicians assess fracture risk, using routine radiographs to estimate bone mineral density (BMD) with accuracy comparable to standard scans.
AI Matches Gold-Standard Imaging
In this 6-year follow-up study, researchers analysed 540 adults who had both dual-energy X-ray absorptiometry (DXA) and standard spine radiographs. An AI tool was used to estimate BMD directly from routine imaging.
The results showed excellent agreement between AI-derived and DXA-derived BMD measurements, with strong correlation and high diagnostic accuracy for osteoporosis. AI-based T-scores achieved an area under the curve (AUC) of 0.959 and an overall accuracy of 90% in identifying osteoporosis.
Predicting Real-World Fractures
Beyond diagnosis, the study examined whether AI-derived BMD could predict fractures over time. During follow-up, participants experienced vertebral, hip, and other fractures.
AI-derived BMD performed comparably to DXA in predicting vertebral and hip fractures, and notably outperformed DXA in predicting overall fracture risk across all sites. Lower AI-derived BMD was significantly associated with increased fracture risk, reinforcing its clinical relevance.
Unlocking Opportunistic Screening
One of the most compelling implications is the potential for opportunistic screening. Unlike DXA, which requires dedicated equipment, routine radiographs are widely available and already performed for other indications.
By leveraging AI, clinicians could identify patients at high risk of osteoporosis and fractures without additional imaging, improving early detection—particularly in settings where DXA access is limited.
A Shift in Osteoporosis Detection
These findings suggest a shift toward more accessible, population-level screening strategies. If validated further, AI-enabled analysis of routine imaging could help close gaps in osteoporosis diagnosis and ensure earlier intervention.
The takeaway is simple: the scans clinicians already have may hold far more value than currently realised—if AI is used to unlock it.
Reference
Hsu JY et al. Can AI-Derived BMD from Routine Radiographs Predict Fracture Risk? A 6-Year Follow-Up Study. Abstract OC24. WCO-IOF-ESCEO, 16–19th April.
Featured Image: Anton on Adobe Stock
