Abstract:
Over the course of a doctoral project, Duncan Bakke completed a body of work
on improving and validating gait modelling, and applying those principles to human
gait studies designed to further explore (and develop novel interventions to
help rehabilitate) post-stroke hemiparetic gait.
It was demonstrated that the MAPClient Lower Limb Scaling Workflow had
higher inter-researcher repeatability than commonly-used linear scaling techniques.
This workflow was used to create all gait models reported in this thesis.
It was demonstrated that the omission of marker clusters on the thigh during
gait capture and inverse kinematics do not alter the resultant joint angle estimations
by a larger magnitude than the uncertainty that the soft tissue artefact of the
thigh introduces. Thigh markers went unused (or unweighted) in all kinematic
analyses reported in this thesis.
A novel haptic biofeedback system based on real-time estimation of peak ani
kle moment was proved to be effective in increasing ankle push-off magnitude in
an able-bodied population.
The same feedback system was pilot-trialled with post-stroke participants,
with results suggesting its use is justified in augmenting post-stroke gait retraining.
Post-stroke gait parameters measured at both self-selected and fast speeds
were used to train unsupervised machine learning categorisation tools called selforganising
maps (SOMs), which correlated with various traditional measures in
a varying manner. One such SOM successfully categorised all post-stroke participants
who attempted high-magnitude ankle push-off feedback thresholds; this
SOM was based on the average knee moment of the non-paretic limb at self-selected
speed.
A novel method of generating synthetic gait data using principal component
analysis (PCA) and convex combination of n-dimensional samples was developed
to use alongside the machine learning approach above.