Deep-learning plus chest x-ray aid osteoporosis management

Sunday, November 29 | 1:00 p.m.- 2:00 p.m. | S4-SSMK01-9 | Room E450A

A deep-learning system trained on a large retrospective dataset may help predict osteoporosis on chest x-rays, according to Australian researchers.

The study findings suggest a way to "close the treatment gap in management of osteoporosis," according to presenter Minh-Son To, MD, PhD, of Flinders University in Adelaide, Australia, and colleagues.

Osteoporosis affects one million Australians and is linked to increased risk of minimal "fragility" fractures, the group explained. It developed a DL model for estimating bone mineral density on plain chest x-rays, matching data from 13,850 thousand bone mineral density studies to 71,560 chest x-rays performed between 2000 and 2020 at South Australia Medical Imaging (the region's public provider of imaging services).

The researchers manually extracted aggregate lumbar spine bone mineral density T-scores and used them to annotate the chest x-rays; this data were split into training, validation and test sets. They then evaluated the prediction performance of the DL model for osteoporosis by calculating sensitivity, specificity and receiver operating characteristic (ROC) curve.

Overall, the group found that the DL model's predicted T-scores strongly correlated with the actual T-scores. Attend this session to learn more about its sensitivity and specificity.

Page 1 of 2
Next Page