Decoding rejuvenating effects of mechanical loading on skeletal aging using in vivo μCT imaging and deep learning.

Throughout the process of aging, dynamic changes of bone material, micro- and macro-architecture result in a loss of strength and therefore in an increased likelihood of fragility fractures. To date, precise contributions of age-related changes in bone (re)modeling and (de)mineralization dynamics to this fragility increase are not completely understood. Here, we present an image-based deep learning approach to quantitatively describe the effects of short-term aging and adaptive response to cyclic loading applied to proximal mouse tibiae and fibulae. Our approach allowed us to perform an end-to-end age prediction based on μCT imaging to determine the dynamic biological process of aging during a two week period, therefore permitting short-term bone aging analysis with 95% accuracy in predicting time points. In a second application, our deep learning analysis reveals that two weeks of in vivo mechanical loading are associated with an underlying rejuvenating effect of 5 days. Additionally, by quantitatively analyzing the learning process, we could, for the first time, identify the localization of the age-relevant encoded information and demonstrate 89% load-induced similarity of these locations in the loaded tibia with younger control bones. These data therefore suggest that our method enables identifying a general prognostic phenotype of a certain skeletal age as well as a temporal and localized loading-treatment effect on this apparent skeletal age for the studied mouse tibia and fibula. Future translational applications of this method may provide an improved decision-support method for osteoporosis treatment at relatively low cost. STATEMENT OF SIGNIFICANCE: Bone is a highly complex and dynamic structure that undergoes changes during the course of aging as well as in response to external stimuli, such as loading. Automatic assessment of "age" and "state" of the bone may lead to early prognosis of deceases such as osteoporosis and enables evaluating the effects of certain treatments. Here, we present an artificial intelligence-based method capable of automatically predicting the skeletal age from μCT images with 95% accuracy. Additionally, we utilize it to demonstrate the rejuvenation effects of in-vivo loading treatment on bones. We further, for the first time, break down aging-related local changes in bone by quantitatively analyzing "what the age assessment model has learned" and use this information to investigate the structural details of rejuvenation process.