An automatic deep learning-based bone mineral density measurement method using X-ray images of children.

Background: Osteoporosis is a common bone disease characterized by low bone mineral density (BMD). Low BMD screening and early interventions during childhood can significantly decrease osteoporosis risk in adulthood. However, in clinical settings, the applicability of dual-energy X-ray absorptiometry (DXA), a technique to measure standard area BMD (aBMD), cannot adequately meet the diagnostic needs of the majority of the Chinese population. We aimed to achieve a comprehensive evaluation in clinical settings by taking a single X-ray image, which, in conjunction with the use of equivalent step phantoms, can assess bone age or injuries (such as sprains, fractures, or breaks) in the wrist while also measuring aBMD in the forearm, to further evaluate growth and development. Methods: In the present study, we used routine X-ray images of the hand and forearm to measure aBMD with step phantom. First, based on the X-ray images, the regions of interest (ROIs) and step phantom used in clinical settings were automatically located and segmented; then, their average grayscale values were calculated. Second, after fitting the linear calibration relationship between the equivalent phantom thickness and grayscale value of the phantom, the effect of soft tissue on aBMD measurement was eliminated using a deep learning method. Finally, aBMD was measured. Results: Our developed method was validated on 500 X-ray images taken at the clinic and compared with DXA-based aBMD measurements. Experiments revealed that the average correlation coefficient was 0.836. Conclusions: The proposed method is an automatic method for measuring aBMD in children by utilizing X-ray images of hand and forearm. Furthermore, our findings suggest the effectiveness of the developed method, which provides a comparable performance to that of clinicians.