Monday, November 27 | 11:30 a.m.-11:40 a.m. | M4-SSMK03-4 | Room E450A
A deep-learning AI model will be presented in this session that can predict bone mineral density T-scores from chest x-rays.
Presenter Yoichi Sato, MD, of Tokyo Shinjuku Medical Center in Japan, and colleagues first developed the model using patient dual-energy x-ray absorptiometry (DEXA) T-scores (bone mineral density) and chest x-rays as input. They then trained the algorithm on a data set of 47,150 x-rays (23,151 patients) and validated it on an external data set of 2,914 radiographs (1,515 patients).
The model achieved 79% accuracy, 96.6% sensitivity, and 34.1% specificity in predicting T-scores ≤ -1.0 (normal) from the x-rays, while T-score predictions ≤ -2.5 (osteoporosis) demonstrated 79.7% accuracy, 77.1% sensitivity, and 80.4% specificity, according to the findings.
“If this deep-learning model becomes available as a medical device, chest x-rays taken in a variety of settings, such as medical examinations, checkups, and hospitals, could be used to screen for osteoporosis,” Sato and colleagues suggest.
