Deep Learning Can ID Low Bone Mineral Density on CT Scans

TUESDAY, Nov. 18, 2025 (HealthDay News) -- Deep learning-based automated opportunistic osteoporosis screening can identify patients with low bone mineral density undergoing computed tomography (CT), according to a study published online Nov. 11 in Radiology.Malte Westerhoff, Ph.D., from Visage Imaging in Germany, and colleagues developed a deep learning-based method to automatically quantify trabecular attenuation using a three-dimensional region of interest (ROI) of the thoracic and lumbar spine on chest, abdomen, or spine CTs. The analysis included 538,946 CT examinations from 283,499 patients performed on 43 scanner models using six different tube voltages.The researchers found that Hounsfield units (HU) differed by 23 percent at 80 versus 120 kVp, and differences in values were 99 percent. Higher mean trabecular attenuation was seen in young women versus young men (younger than 50 years), which decreased with age; postmenopausal women had a steeper decline. Trabecular attenuation was higher in men than women among those aged older than 50 years. The highest trabecular attenuation was seen in Black patients, followed by Asian patients, with the lowest seen in White patients. For L1, the threshold for diagnosing osteoporosis was 80 HU."Our study offers proof that existing medical images done for other reasons can be repurposed and used to reliably identify bone loss, such as in osteoporosis," senior author Miriam A. Bredella, M.D., from NYU Langone Health in New York City, said in a statement.Several authors are employed by Visage Imaging.Abstract/Full Text.Sign up for our weekly HealthDay newsletter