DXA Image Based Deep Learning of Elastic Modulus of Human Trabcular Bone in Proximal Femur

Osteoporosis is a skeletal disorder, characterized by low bone mineral density (BMD) and bone quality deterioration, which may lead to bone fragility fractures. It is estimated that 200 million people have osteoporosis with 2% to 8% of men and 9% to 38% of women are affected. Various methodologies are currently used to predict bone fragility fractures. Among these modalities, DXA is the most accessible and affordable means that could be used to determine both BMD and microstructural features of bone. The goal of this study is to determine whether DXA image-based deep learning (DL) model could predict the elastic modulus of human trabecular bone. To achieve the goal, 591 trabecular bone cubes were digitally dissected out from six human cadaveric proximal femurs. Simulated DXA images were generated from each trabecular cube as input, whereas the elastic modulus that was estimated using FEM simulations as output for training the DL model. The results of this study verified the efficacy of the DL model in predicting the elastic modulus of trabecular bone (R=0.95) and showed that the prediction accuracy of the DL model was dependent on the sample size and the number of input DXA images. In addition, the DL model was compared to a multiple linear regression model where six histomorphometric parameters (i.e., BV/TV, BS, Tb.Th, DA, Conn.D, and SMI) were used as independent variables and the elastic modulus as dependent variable. A strong correlation (R=0.93) was observed between the two models in predicting the elastic modulus of trabecular bone, suggesting that the DXA image based DL model could capture the effect of microstructural features on the elastic modulus of trabecular bone as did the regression model. The outcome of this study indicates a potential of using DL techniques for predicting the mechanical properties of trabecular bone solely from DXA images.