Abstract:
This thesis is divided into three distinct contributions to modelling in orthopaedics. Firstly,
modelling of bone cell diffusion in a scaffold using Fick’s law is presented that is used to
train support vector machine regression and convolutional neural network models. The
machine learning predictions are compared against experimental observations. Secondly,
machine learning is applied to the whole bone from a dataset derived from the Victorian
Institute for Forensic Medicine. Twenty-five HR-pQCT scans of human proximal femurs are
solved in Abaqus to predict bone strains. Using additional synthetic data derived from these
models a neural network and partial least squares regression model are trained on the
mechanical simulations. The methods are compared for predictions and what insights can be
revealed. Thirdly, a model for improving estimates of boundary conditions is presented using
a sheep animal model and wearable sensors. Using measures of sheep activity we present a
daily loading stimulus measure that can be used to integrate sheep-specific behaviour with
finite element simulations. This has useful implications in animal models where orthopaedic
evaluations are often performed.