Comprehensive Summary
This study developed an artificial intelligence system to estimate lumbar and femoral bone mineral density (BMD) from standard lumbar X-rays, offering a low-cost alternative to DXA scanning for osteoporosis screening. Using data from 1,454 participants in Japan’s ROAD cohort, a transformer-based neural network predicted BMD values with DXA as the reference. The model achieved strong performance, with mean absolute errors of 0.076 g/cm² for lumbar and 0.071 g/cm² for femoral BMD, and r values of r = 0.89 and r = 0.74, respectively. It identified osteopenia and osteoporosis with accuracies of 85% and 79% and AUCs up to 0.95. Performance was the same across BMI and sex but declined slightly in patients with severe spinal degeneration. Although limited by single-center data and no external validation, the findings suggest that AI applied to routine lumbar X-rays could enable accessible, opportunistic osteoporosis screening in everyday clinical care.
Outcomes and Implications
For the medical community, this study highlights how artificial intelligence could make osteoporosis screening more accessible by using standard lumbar X-rays to estimate both lumbar and femoral BMD. This system could allow earlier detection of low bone mass in primary care, supporting preventive treatment. The model’s strong accuracy and consistency across demographic groups indicate clinical potential, though its validation is currently limited to a single Japanese cohort. Broader testing across diverse populations and imaging systems will be essential before implementation. If validated, this technology could reduce screening costs, integrate seamlessly into routine imaging workflows, and significantly expand osteoporosis detection without additional radiation exposure.