Improved femoral neck fracture predictions using anisotropic failure criteria models

  • Pithioux Martine
  • Chabrand Patrick
  • Hochard Christian
  • Champsaur Pierre

  • Femoral neck fractures
  • Failure criteria
  • Finite element analysis
  • Experimental tests
  • Transversely isotropic material


Finite element models are widely used to assess long bone strength, implant stability and other clinical problems. In most of the models presented so far in the literature, the bone is taken to be isotropic, and the occurrence of failure is predicted by defining a threshold von Mises stress. However, human bone is found to show orthotropic behavior. Studies so far have focused only on the use of anisotropic criteria in orthotropic models designed to predict the occurrence of human femur failure. The aim of this study was therefore to investigate how specific finite element models for human femora combined with composite failure theories could be used to improve failure predictions in vitro. For this purpose, nine human proximal femora were tested mechanically up to failure under the loading conditions present during the one-leg stance phase in walking. Specific finite element models using various materials to represent the bone were generated for each femur. First, the bone material was modeled in the form of an isotropic brittle material, and the von Mises criterion was used to predict the occurrence of fracture. Second, the bone was modeled as a transversely isotropic brittle material with asymmetric strength characteristics, and the occurrence of fracture was predicted using the Hill and the Tsai-Wu criteria. The results obtained here show that the transversely isotropic model combined with Tsai-Wu and Hill criteria accurately predicted the fracture load (values of R(2) = 0.94 and SEE = 10.3% were obtained with the Tsai-Wu criteria and R(2) = 0.82 and SEE = 22.9% were obtained with the Hill criteria), while the isotropic model combined with the von Mises criterion overestimated the fracture load, although a good correlation was generally observed with the experimental results (R(2) = 0.77, SEE = 30.6%).