Predicting Femoral Strain Using Surrogate Modelling

Author: Thomas Rundle

Rundle, Thomas, 2023 Predicting Femoral Strain Using Surrogate Modelling, 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.


Quantification of femoral strain in real-time is valuable for a range of biomedical fields as it enables rapid assessment of fracture risk. Amongst individuals living with spinal cord injury, bone fracture during rehabilitation and exercise poses a particularly high risk given diminished bone mass. Further, the lack of sensory feedback can result in injuries untreated and lead to health implications.

Currently, the finite element (FE) method is used to predict femoral strains in response to applied loads. Although the FE method has been validated for many models of bone mechanics, it is time-consuming, requires high-level training to operate, and requires extensive model development for each new application (e.g., patient). This study proposes a method which uses surrogate modelling of legacy datasets to predict femoral strain in response to novel and, in principle, arbitrary applied loading.

Four techniques were investigated: multi-linear regression (MLR), cubic splining, Superposition Principle Method (SPM), and Kriging. Surrogate models were created in MATLAB (Mathworks, USA), and were used to predict femoral strains in response to various loads.

An initial linear FE models showed SPM predicted FE-modelled strains with zero error. The MLR and cubic splining techniques were both effective, with nRMSE values of <0.05% and <0.08% respectively. Kriging was inaccurate, with nRMSE >15%. All techniques were computationally tractable, but MLR was slowest taking ~14.9 seconds while splining was fastest taking ~1.50 seconds. When applied to the non-linear model, SPM was still accurate, with nRMSE of ~5% and CPU time <20 seconds.

The SPM is the recommended surrogate modelling technique for applications requiring near-real-time femoral strain quantification. Despite being a lesser known and under-developed method, it provided exact strains in a linear model, and highly accurate ones in a non-linear model in a timely manner. Other methods were found to be less favourable, however their lack of testing in a non-linear environment should be considered. Through code optimization, it is expected that SPM could run in real-time.

Keywords: Surrogate modeling, FEA, strain, fracture risk, biomedical engineering, bone fracture

Subject: Engineering thesis

Thesis type: Masters
Completed: 2023
School: College of Science and Engineering
Supervisor: Mark Taylor