Calculation of femoral strain during normal activities using efficient computational methods

Author: Hamed Ziaei Poor

Ziaei Poor, Hamed, 2019 Calculation of femoral strain during normal activities using efficient computational methods, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


Efficient calculation of femoral strain is important for various biomechanical applications, such as predicting femoral strains over multiple physical activities; improving the design of implantable devices; and assessing the risk of femoral fracture. The finite-element method has been used largely for the calculation of femoral strain. However, generating and solving the models using common image-based procedures make it practically impossible, particularly for large studies of bone which need to run the thousands of simulations to explore the interdependence between femur anatomy, bone quality, strain, and motor task using lots of point data, meaning that FE models are extensively large and computationally too expensive to run. Surrogate modelling techniques are used successfully for various applications in biomechanics however the characteristics of these methods need to be investigated further.

This thesis proposes a novel technique for the efficient calculation of femoral strain over multiple activities and individuals. This is achieved by taking three different steps: (I) assessing the ability of Multivariate Linear Regression (MLR) to predict the femoral strain field in a single participant while executing different normal activities; (II) developing a training-free method based on the Superposition Principle Method (SPM) and comparing its performance against that of three popular surrogate models; and (III) evaluating the feasibility of using Principle of superposition for prediction of femoral strain within a cohort of patients. To achieve this, Superposition Principle Method needs to be integrated with an Active Shape and Appearance Model, allowing to predict femoral strain by capturing the variation in femur geometry, bone distribution, and thereby enabling population-based studies into multiple subjects and motor tasks.

The MLR model provided a viable solution for the rapid calculation of full femoral strain fields by estimating femoral strain for a full activity cycle in 13 seconds compared with 55 minutes for FEM, enabling large statistical analyses. The SPM method provided the lowest error (RMSE = 40 me), the fastest model construction time (3.2 h) and the second-fastest prediction time per activity (36 s) after the MLR method. Finally, the integration of SPM combined with statistical shape and appearance models enabled efficacy in the prediction of femoral strain for an arbitrarily selected instance within the population and motor task. The peak error was 1.4 – 4.9%, in agreement with the error (i.e., 4.2 – 8.3% of peak strain) in current FE technologies based on CT images for predicting femoral strain. When the performance of the model was examined for three randomly selected participants which were not used previously for building the model, a good correlation was observed between the FE and SPM2 strain calculations (RMSE < 10% of peak strain, R2 > 0.86). This was in agreement with previous studies (RMSE 11 – 15 % of peak strain, R2 > 0.88). The major source of error was related to the reconstruction of three independent femurs from the Active Appearance Model (RMSE = 488με, R2 = 0.95) compared with the error originated from Active Shape Model (RMSE = 261με, R2 = 0.97). In conclusion, the combination of the Superposition principle and Principal component analysis is an effective solution for a running population-based simulation of femoral mechanics during activity. However, active appearance models need further improvement to be used for large biomechanical analysis of intact or implanted bone.

Keywords: Calculation of Femoral Strain; Efficient Computational Methods; Superposition Principle Method; Different Surrogate Modeling Techniques; Active Shape and Appearance Models, Population-based Study

Subject: Computational Modelling thesis

Thesis type: Doctor of Philosophy
Completed: 2019
School: College of Science and Engineering
Supervisor: Prof. Mark Taylor