Every day, thousands of dual-energy X-ray absorptiometry (DXA) scans are performed to assess bone health. Clinicians use these images to measure bone mineral density, helping to identify individuals at risk of osteoporosis and fracture. But what if these scans are telling a much bigger story? Beyond bone, DXA images capture information about the body's muscles and fat distribution, providing a snapshot of musculoskeletal and metabolic health that is rarely explored. With recent advances in artificial intelligence (AI), we can begin to uncover these hidden signals and reveal insights that extend far beyond bone.

Despite DXA machines being capable of assessing body composition, including appendicular lean mass (ALM) the required whole-body DXA scans are not currently reimbursed by Medicare in Australia and are therefore not acquired in clinical practice. Unsurprisingly, sarcopenia, comprising low ALM and strength, is rarely assessed clinically despite it being a key risk factor for falls and fracture. Similarly, magnetic resonance imaging (MRI)-derived measures, such as fat-free muscle volume (FFMV) and muscle fat infiltration (MFI), provide a more comprehensive assessment of muscle quantity and quality. These measures have been shown to be important determinants of physical function, frailty, falls risk, and other adverse health outcomes. However, the cost and limited availability of MRI restrict its use in large-scale screening and routine clinical care.

This presentation will describe recent advances in the application of AI to routine hip and spine DXA images for the assessment of cardiometabolic biomarkers, specifically muscle mass and quality, as well as other emerging phenotypes (e.g., visceral fat, AAC). Leveraging large imaging datasets and deep learning methods, we have developed models capable of estimating MRI-derived measures of muscle composition directly from DXA scans. These approaches enable the extraction of additional important clinical information from DXA images already acquired in millions of people worldwide for osteoporosis screening, without additional imaging burden, cost, or radiation exposure.

Preliminary results from large-scale population studies demonstrate that these AI-derived muscle quality measures show strong agreement with reference-standard MRI assessments and provide complementary information to traditional DXA-derived measures such as BMD in fracture risk assessment in a real-world BMD registry. The presentation will discuss the potential clinical applications of these technologies for identifying individuals at risk of poor musculoskeletal health outcomes, enhancing the value of existing DXA infrastructure.