Background: Osteoporosis remains a disease which is both underdiagnosed and undertreated. It disproportionately affects women, elderly and those in lower socioeconomic groups [1]. We postulated there was a better pathway for screening osteoporosis.
Method: We extracted 71,560 paired chest x-rays and BMD studies from the SAMI database which were performed within 6 months of each other. These paired datapoints were used to train a deep learning model to predict the T-score from chest x-ray.
The model was tested on a subset of the data which was not used in the training.
Results: The T-score was able to be predicted on chest x-ray with an AUC of 0.884. Chest x-ray was able to screen for osteoporosis with an 80.9% sensitivity, 82.0% specificity, 50.0% PPV and 95.2% NPV.
Discussion: A good screening test is one which includes a large cohort, is acceptable to the population and earlier intervention will alter disease morbidity/mortality.
Currently 200,000 screening BMD studies are performed annually. Chest x-rays are the most common radiological procedure with 1,800,000+ performed annually (large cohort) [2].
Radiologists are good at reporting the majority chest x-ray findings but poor when evaluating for osteoporosis [3]. This is due to the multiple imaging parameters which impairs consistent analysis. Deep learning has demonstrated the ability to synthesise the different parameters to provide an estimation of T-score. The program does not alter the chest x-ray patient journey and therefore is an acceptable adjunct i.e. no extra radiation or time.
The model does not replace BMD but identifies patients who would benefit from further investigation with a formal BMD study. We believe it can increase screening of osteoporosis by 9x.
Of the highlighted patients, there will be 50% osteoporosis, 40% osteopaenia and 10% normal findings on follow-up BMD study. Are endocrinologists ready for an increase in osteoporosis diagnosis?