Sunday, November 29 | 1:00 p.m.-2:00 p.m. | S4-SSMK01-9 | Room E450A
An AI algorithm trained to predict bone mineral density from routine chest x-rays demonstrates promising performance for opportunistic osteoporosis screening, according to research in this Sunday session.
Minh-Son To, MD, PhD, of Flinders University in South Australia, and colleagues matched 71,560 chest x-rays to 13,850 bone mineral density (BMD) studies performed between 2000 and 2020 and trained a deep-learning model on an EfficientNetV2-M architecture to predict lumbar spine T-scores directly from chest radiographs.
Predicted T-scores were strongly correlated with actual measurements, with a Pearson correlation coefficient of 0.77. For detection of osteoporosis, defined as a T-score of -2.5 or lower, the model achieved an area under the ROC curve of 0.90. At a prediction threshold of -1.5, sensitivity and specificity for osteoporosis detection were 81% and 82%.
Osteoporosis affects an estimated one million Australians and is associated with elevated risk of fragility fractures, the researchers noted. The majority of patients who present to hospital with a fragility fracture are neither assessed nor appropriately managed for osteoporosis, a treatment gap the team said opportunistic chest x-ray screening could help close.
“Our deep-learning system trained on a large retrospective dataset shows promising performance for prediction of osteoporosis on chest X-rays,” the authors wrote.
Catch this Sunday session for more on the model’s performance and what the team sees as next steps.



