A new meshless approach for subject-specific strain prediction in long bones: Evaluation of accuracy

Background: The Finite Element Method is at present the method of choice for strain prediction in bones from Computed Tomography data. However, accurate methods rely on the correct topological representation of the bone surface, which requires a massive operator effort, thus restricting their applicability to clinical practice. Meshless methods, which do not rely on a pre-defined topological discretisation of the domain, might greatly improve the numerical process automation, but currently their application to biomechanics is negligible. Methods: A meshless implementation of an innovative numerical approach based on a direct discrete formulation of physical laws, the Cell Method, was developed to predict strains in a cadaver femur from Computed Tomography data. The model accuracy was estimated by comparing the predicted strains with those experimentally measured on the same specimen in a previous study. As a reference, the results were compared to those obtained with a state-of-the-art finite element model. Findings: The Cell Method meshless model predicted strains highly correlated with the experimental measurements (R² = 0.85) with a good global accuracy (RMSE = 15.6%). The model performed slightly worse than the finite element one, but this was probably due to the need to sub-sample the original data, and the lower order of the interpolation used (linear vs parabolic). Interpretation: Although there is surely room for improvement, the accuracy already obtained with this meshless implementation of the Cell Method makes it a good candidate for some clinical applications, especially considering the full automation of the method, which does not require any data pre-processing.