gms | German Medical Science

Objectives: There is still a high prominence of healing issues in bone fractures and thus understanding and predicting why healing failed is crucial. Perren’s strain theory demonstrated that bone fracture healing progresses in response to the mechanical stimuli at the fracture site. From this, mechanoregulatory computer simulations have been developed to predict fracture healing and give insight into different healing phenomena. However, these models remain unvalidated against a ground truth measurement. In this study, specimen specific prognostic fracture healing simulations are compared with corresponding in vivo sensor data for the first time to validate these models.

Methods: Data from a previously completed ovine tibial osteotomy study was used and contained 16 sheep. A custom plate designed for controlled axial motion had a sensor that continuously measured the displacement in the healing region. Computed tomography derived finite element (FE) models were created of the construct and was modeled with the implant. Prognostic fracture healing simulations were performed for 56 days in using an iterative algorithm. The FE simulations calculated a virtual displacement in the healing region, analogous to the embedded in vivo sensor. Furthermore, a percentage of healing was determined as the FE-based virtual torsional rigidity (VTR) of the ipsilateral bone, normalized to the contralateral side. For each signal (in vivo sensor, virtual sensor, and VTR), a healing time was determined by reaching a threshold of <0.1 mm displacement or >90% VTR and maintaining a steady state with <5% change from the signal 3 days prior sustained for 5 days.

Results and conclusion: The prognostic fracture healing simulations calculated healing as a change in the tissue growth and mineralization, showing the onset of bridging with a steady state effect after healing has occurred (Figure 1a [Fig. 1]). There were no significant differences between the in vivo measured and in silico predicted healing times as determined by a one-way repeated measures ANOVA (p=0.22; Figure 1b [Fig. 1]) with median healing times of 28.5, 27.5, and 28.8 days for the in vivo displacement sensor, virtual displacement sensor and VTR tests, respectively.

This work was the first time a prognostic fracture healing simulation was performed on a specimen specific model with accompanying in vivo sensor data. The FE simulations correctly predicted healing time, indicated by the lack of differences between the in vivo and in silico results. These demonstrate the power of these predictive models that could be used in the future to assist in the design of implants, determining weight bearing protocols, and assessing aseptic nonunion risk.