Approximating femoral neck bone mineral density from hand, knee, and pelvis X-rays using deep learning

Abstract

Background: A tool trained to learn the complex features of bone and soft tissue attenuation to estimate bone mineral density (BMD) at the femoral neck from standard hand, knee, and pelvis X-rays has the potential to opportunistically screen for low BMD in individuals that undergo such X-rays for any clinical indication, which in turn could empower patients and their providers to initiate preventative treatment. Methods: A retrospective study of the Osteoarthritis Initiative (OAI) dataset consisting of hand, knee, and pelvis X-rays and corresponding dual-energy X-ray absorptiometry (DXA)-derived femoral neck BMD (examinations done between 2008 to 2010) from 553 unique patients with osteoarthritis (OA) (51% male), aged between 48 to 83 years old. Participants were divided into training and test splits using a stratified random sampling procedure to ensure equal distribution of sex and age decade. A deep convolutional neural network (CNN) was trained to learn visual features from raw X-ray images, which were then combined with sex and age of the patients to estimate their femoral neck BMD. Agreement between methods at estimating BMD was assessed with Passing-Bablok regression and Bland-Altman analyses. Agreement between methods at classifying low BMD (T-score 75% in both females and in males. It is also shown that both the X-ray and co-variate data equally contribute to the model performance. Conclusions: These results indicate that low BMD at the femoral neck can be opportunistically screened from routinely acquired X-rays of the hand, knee, or pelvis, i.e., even when the femoral neck is not included in the field of view.