Abstract:
Cerebral palsy is a well-recognised neurodevelopmental condition beginning in
early childhood and persisting throughout the lifespan of the affected individual.
Neuromuscular disorders and musculoskeletal deformities result in gait abnormalities for
children with cerebral palsy. Equinus gait, characterised by excess plantarflexion, is one
of the most common abnormal gait patterns and receives much attention from surgeons
and therapists. Three-dimensional gait analysis plays an essential role in cerebral palsy
management. However, the conventional optoelectronic three-dimensional gait analysis
system is expensive and not user-friendly to individuals without the necessary expertise,
hampering its application in clinical practice. In addition, in-depth knowledge regarding
ankle muscle activation and force generation ability during ambulation can help clinical
practitioners develop suitable treatment strategies for children with cerebral palsy and
equinus gait. Therefore, the first aim of this thesis was to determine the validity and
reliability of gait parameters assessed by Kinect-based markerless gait analysis systems
for children with cerebral palsy. The second aim of this thesis was to determine the effect
of involving more personalised musculoskeletal information on ankle muscle force
prediction for children with cerebral palsy and equinus gait.
The work undertaken in this thesis includes five experimental studies. The first three
studies determined the validity, and the reliability, of cost-effective markerless threedimensional
gait analysis systems based on Kinect sensors when evaluating crucial gait
parameters for children with cerebral palsy. The last two studies investigated the effect
of involving pre-calculated joint angles for musculoskeletal model scaling, and applying
image-based musculoskeletal models on ankle muscle force prediction, for children with
cerebral palsy and equinus gait.
The results of the first two studies revealed that the single Kinect V2-based threedimensional
gait analysis system was unable to provide accurate joint kinematic
estimations. Its validity could be improved by implementing calibration algorithms,
including the linear regression-based method and the long short-term memory recurrent
neural network-based approach. Also, some hip and knee sagittal kinematic variables
(e.g., maximal/minimal hip flexion/extension angles and maximal knee flexion/extension
angles) demonstrated good inter-day reliability. Most of the selected gait spatiotemporal variables (e.g., step length/width/time, stride length/time, gait speed) captured by the
single Kinect V2-based gait analysis system were accurate and reliable. Furthermore, this
system could accurately assess the minimal margin of stability, an essential indicator of
fall risk in previous studies. Although the single Kinect V2-based gait analysis system
could not provide subtle gait metric evaluations, it still had the potential to be utilised as
a portable alternative for some coarse analyses (e.g., gait classification and screening),
assess gait stability and monitor gait progressions. In the third study, a dual Azure Kinect
gait analysis system was developed based on the point cloud registration algorithm to
provide full-scene motion capture. The results demonstrated that the dual Azure Kinect
system could still not offer an accurate kinematic assessment.
In the fourth study, children with CP and equinus gait were recruited to conduct the
three-dimensional gait analysis. Each participant’s ankle muscle forces were estimated
by static optimisation and EMG-assist modelling approaches with two scaled
musculoskeletal models. The results demonstrated that imposing pre-calculated ankle
angles during musculoskeletal model scaling could improve the agreement between ankle
dorsi/plantarflexion angles calculated by inverse kinematics and referential direct
kinematics. Implementing pre-calculated joint angles during musculoskeletal model
scaling could also affect muscle force modelling performance. For the static optimisation
approach, the tibialis anterior muscle force prediction was more likely to be influenced.
For the EMG-assisted modelling, the gastrocnemius muscle force prediction was more
likely to be affected. Scaling with pre-calculated joint angles was recommended due to
the better consistency between joint angles estimated by direct kinematics and inverse
kinematics, facilitating communication among researchers from different research
disciplines. In the fifth study, patient-specific musculoskeletal models were developed for
the participants by using MRI scan images of each participant. The results demonstrated
that involving personalised joint rotation axes derived from MRIs influenced the inverse
kinematic calculation and muscle force prediction, showing the importance of involving
personalised anatomical information derived from medical images in
neuromusculoskeletal modelling for children with cerebral palsy and equinus gait.
In summary, this thesis determined the validity and reliability of two Kinect-based
cost-effective gait analysis systems for children with cerebral palsy. The feasibility of
utilising Kinect-based three-dimensional gait analysis in screening gait patterns,
monitoring gait progressions, and evaluating gait stability is established. This thesis also
provided methodological evidence for clinicians and researchers who intend to acquire more personalised biomechanical simulation results for children with cerebral palsy and
equinus gait.