Supersampling Enables Accurate Microstructural Bone Adaptation Simulations in Human in vivo HR-pQCT Images

In silico trials of treatments in a virtual physiological human (VPH) would revolutionize research in the biomedical field. Hallmarks of bone disease and treatments can already be simulated in pre-clinical models and in ex vivo data of humans using microstructural bone adaptation simulations. The increasing availability of in vivo high resolution peripheral quantitative computed tomography (HR-pQCT) images provides novel opportunities to validate and ultimately utilize microstructural bone adaptation simulations to improve our understanding of bone diseases and move towards in silico VPH decision support systems for clinicians. In the present study, we investigated if microstructural bone adaptation simulations of in vivo human HR-pQCT images yielded accurate results. Since high-resolution ground truth images cannot be obtained in vivo, we applied an ex vivo approach to study resolution dependence and the effect of upscaling on morphometric accuracy. To address simulation initialisation issues, we developed an input regularisation approach to reduce initialisation shocks observed in microstructural bone adaptation simulations and evaluated upscaling as a way to improve the accuracy of model inputs. Finally, we compared our ex vivo results to simulations run on in vivo images to investigate whether in vivo image artefacts further affect simulation outcomes.