CT imaging of the chest, abdomen, and spine to detect conditions such as kidney stones or pulmonary lesions can also be used to identify early signs of osteoporosis using an artificial intelligence (AI) tool, a new study shows.

“Clinicians should consider the value of opportunistic CT screening for osteoporosis as an adjunct to traditional methods, such as dual-energy x-ray absorptiometry [DEXA],” said Miriam Bredella, MD, Bernard and Irene Schwartz professor of radiology and associate dean for translational science and director of the Clinical and Translational Science Institute at the NYU Grossman School of Medicine, New York.

Data from the new study show that AI-driven analysis of routine CT scans can help identify patients at risk of fracture, facilitate timely interventions, and potentially reduce costs and improve patient outcomes, Bredella said in an interview.

In the study recently published in Radiology, the researchers described the development and testing of an AI algorithm to screen for osteoporosis in individuals undergoing CT for other purposes. 

The algorithm was based on reported prevalence of osteoporosis by the World Health Organization and assessed bone mineral density for all major lumbar and thoracic vertebra in the body according to thresholds for age, gender, and race and ethnicity.

The researchers tested the algorithm on 538,946 CT examinations in a study population of 283,499 patients seen at NYU Langone. The mean age of the participants was 65 years, 51.2% were female, and 55.5% were White. The CT scans reflected a range of CT scanning protocols. 

A total of 95,465 women aged 60-69 years from the study population underwent DEXA screening at the study institution; the prevalence of osteoporosis in this group was 23.74%. An analysis of a subset of 6373 of these patients who had had a CT scan within 12 months of their DEXA screening revealed a nearly identical osteoporosis prevalence of 23.65%. 

The researchers identified several trends in bone loss across a diverse population of patients using the AI algorithm, and radiologists verified the findings.

Notably, the results showed higher bone density in women younger than 50 years compared to male peers. With age, the difference decreased, and bone density declined among postmenopausal women. When assessed by race, bone density was highest in Black patients, followed by Asian and White patients. 

Endocrinology Implications

“We performed this study because osteoporosis is underdiagnosed and undertreated, largely due to limited referrals for traditional DEXA screening,” Bredella told Medscape Medical News. 

“An important part of the study was also to use different scanner models and protocols to address the variability in bone density values caused by diverse imaging parameters,” she noted. “Our study is the first to develop a robust statistical method for calibrating measurements across different scanners and protocols, ensuring consistency and accuracy in diagnosing osteoporosis,” she said.

Study limitations include the lack of generalizability given the use of data only from the northeastern United States, said Bredella. 

“Also, we used a 3D method specifically designed to identify the area of lowest density within a vertebral body. As a result, the thresholds established in this study are only applicable when using the same 3D technique, and they should not be directly applied with 2D methods or alternative [region of interest] placements without adjustments,” she added. 

Not Ready for Prime Time 

DEXA technology remains underutilized among older adults, including older women, and osteoporosis remains underdiagnosed, even in high-income countries, said Kristine E. Ensrud, MD, professor of epidemiology, community health, and medicine at the University of Minnesota, Minneapolis, in an interview.

The current study findings were not surprising, as the study is one of many that have evaluated the use of CT scans obtained for other indications as screening tools for osteoporosis, said Ensrud, who was not involved in the study 

The cutpoints for greater spinal trabecular attenuation proposed in the current study identified a similar proportion of patients with osteoporosis as was present in the subset of patients who had received areal bone mineral density (aBMD) measurements (about 23%), said Ensrud. However, the researchers did not report the agreement between these two methods of diagnosis, which is a concern and limitation of the results, she noted.

“Important barriers must be addressed before widespread implementation of opportunistic osteoporosis screening using methods like the one proposed in the current study,” she told Medscape Medical News. 

Adding additional software to CT scanners is an expense for health systems, she said, and “whether cutpoints of trabecular attenuation have the same sensitivity and specificity for identifying osteoporosis across different CT scanners is unknown.” 

“A patient who exceeds the cutpoint for greater trabecular attenuation will still need to undergo DEXA testing to obtain aBMD of hip and spine in order to make the diagnosis of osteoporosis and determine whether osteoporosis drug treatment is indicated to prevent fractures,” she emphasized. 

Importantly, identification of osteoporosis on a CT scan taken for another purpose is considered an incidental finding, and many CT scans already report incidental findings, said Ensrud. “Following up on incidental findings typically lands in the work queue of overburdened primary care clinicians who often did not order the CT scan,” she noted. Before CT scans can be widely leveraged to identify osteoporosis, healthcare system workflows are needed that include input and buy-in from all those involved, she added. 

Ultimately, research would be needed to determine whether implementing opportunistic osteoporosis screening in healthcare systems lowers fracture rates or reduces fracture-related healthcare costs, Ensrud said. 

Bredella disclosed no financial conflicts of interest. Several coinvestigators are employees of Visage Inc, the company that co-developed the algorithm. 

Ensrud disclosed no financial conflicts of interest.