This study was performed to investigate the possibility of applying AI techniques in prediction of trabecular bone microarchitecture and mechanical behavior based on low-resolution medical images (i.e., DXA and QCT). Firstly, the feasibility of using both DXA and QCT images to train high-fidelity DL models in prediction of trabecular bone microstructural features and mechanical properties was investigated. Secondly, a probability based generative model was developed to render digital models of trabecular bone, which could be potentially used to train high-fidelity DL models in predicting mechanical properties of trabecular bone. Finally, a deep transfer learning model assisted by the generative model was proposed to train a high-fidelity predictive model of trabecular bone mechanical properties using a small number of real bone samples. The results showed that (1) Both DXA and QCT based DL models had high accuracy in prediction of microstructural features and mechanical properties of trabecular bone cubes (i.e., representative volume element), (2) The generative model developed in this study could only partially match the microarchitectural features and mechanical properties of target real bone samples, (3) Assisted with the generative model of trabecular bone, high-fidelity deep transfer learning models could be trained to predict mechanical properties of trabecular bone using a limited number of real bone samples. The results supported the hypotheses of this study and achieved the proposed objectives. vi