An AI tool could flag patients at risk for osteoporosis or osteopenia by evaluating standard hip and pelvis x-rays, new research showed.

The tool could benefit early intervention for low bone mineral density (BMD), especially in primary care or in settings where DEXA is not always readily available, according to the study investigators. Patients who are flagged with the algorithm, called OsteoSight (Naitive Technologies Limited), would then receive additional screening.

Bone Disorders Underdiagnosed

Although guideline-recommended screening options already do exist for patients at risk for low bone density, as many as 70% of people with osteoporosis remain undiagnosed.

“DEXA is sometimes hard to access,” said Michael C. Willey, MD, an associate professor of Orthopedics and Rehabilitation at the University of Iowa, in Iowa City, Iowa, who presented the new findings at the 2026 annual meeting of the American Academy of Orthopaedic Surgeons. In rural settings, a DEXA machine might not be close by, he said, and patients often must schedule a separate appointment following the initial visit.

“Even though DEXA is recommended for osteoporosis screening, many people do not follow these recommendations because of inconvenience and/or cost,” Willey said.

The problem with skipping screening, he said, is missing someone with low bone density, and the repercussions of low BMD: predisposition to fractures, which often require surgeries, he said.

The new tool, which integrates into the electronic health record, could be another screening option, flagging patients before a DEXA scan. If someone has pain from suspected arthritis or a fall, x-rays are widely available at healthcare offices, not only at the orthopedist but also increasingly with primary care.

“We commonly get x-rays, and that’s a point of care where we can potentially identify people that are high risk for osteoporosis based on their bone density on x-ray,” Willey said.

High Specificity for BMD

In the new study, Willey and his colleagues collected de-identified data from 2458 patients who had visited seven independent US imaging centers and received both a DEXA scan and an x-ray that was analyzed by OsteoSight, which the FDA approved in 2025. The investigators required that the x-rays be of the hip or pelvis and DEXA scans to be from the femoral neck; DEXA T-scores below -1.0 SD (prevalence, 67%) were considered low BMD.

The investigators found the AI tool had an area under the operating characteristic curve of 0.836 (95% CI, 0.819-0.853), a specificity of 0.942 (95% CI, 0.922-0.961), and a sensitivity of 0.448 (95% CI, 0.387-0.500). Observed agreement in repeat measurements was 84%, they reported.

Willey said that he and his co-investigators were happy with the results, and that they were not concerned with the moderate sensitivity. “The AI software is designed to have a high specificity,” he said. “We don’t want to miss people that have osteoporosis with the screening.”

X-Ray Quality Could Be a Factor

The technology could have a place in primary care, according to Kevin Baker, MD, a diagnostic radiologist and assistant professor at Stony Brook Medicine, in Stony Brook, New York, who was not involved in the study. However, the quality of the radiographs could affect the accuracy of the tests, Baker said.

“Part of the contribution to uncertainty regarding bone density on x-rays though is how they were acquired,” he said. The way the technician places the person’s body or equipment during x-rays “can throw off the way that bone density is perceived, so I could see those as confounding factors,” he said.

With this caveat, the widespread availability of x-rays means many people could benefit from this screening tool, especially in primary care, Willey said. “The ultimate goal is to identify people that are high risk for having fractures so that they can be referred on to be treated to improve their bone density and hopefully reduce the incidence of fragility fractures.”

Barbara McHugh, vice president of product management for Naitive Technologies, based in London, England, said the tool is currently being used by “a range of clinical sites” in the US, including private practices and academic medical centers.

Baker reported having no relevant financial conflicts of interest. Willey reported receiving research support from Naitive Technologies to conduct the study.

Meaghan Lee Callaghan is a science journalist based outside of New York City. Her work has appeared in Audubon, Infectious Disease Special Edition, Popular Science, and Scientific American, among other places.