Biomedical Engineering - Theses
Now showing items 1-12 of 48
Muscle and joint functions during walking in individuals with transfermoral amputation
Individuals with unilateral transfemoral amputation depend on compensatory muscle and joint function to generate motion of the lower limbs, which can produce gait asymmetry. Osseointegration is an alternative technique to socket-based prostheses that is used for reducing socket-skin contact problems. However, between-limb differences in joint kinematics and net joint moments may lead to abnormal hip joint contact behavior and muscle function. The aim of this dissertation is to investigate gait compensatory mechanism in individuals with transfemoral amputations fitted with socket (TFA) and bone-anchored prostheses using osseointegrated implants (BAP). In this study, two experimental and computational approaches were used to quantify the contributions of the intact and residual limb’s contralateral muscles to body center of mass acceleration and hip joint contact forces during walking. In the first approach, kinematics and kinetics data were collected from 6 TFAs and 4 BAPs performing over-ground self-selected walking task. In the second approach, a processing framework was employed using OpenSim software and MATLAB API scripting for developing three-dimensional musculoskeletal models and then to predict muscle forces and muscle contribution to waling and hip joint reaction forces. It was found that the intact limb hip muscles contributed more to body center of mass acceleration and hip contact forces than those in the residual limb. The results also suggest that osseointegrated amputees could quantify to decrease the asymmetries in the biomechanical measures between the intact and residual limbs than socket-based prosthesis amputees. The findings of this study would be useful in developing rehabilitation training programs and design of prostheses to improve gait symmetry and mitigate post-operative musculoskeletal pathology.
Cerebral perfusion markers in full-term neonates as a measure of Hypoxic-Ischemic Encephalopathy during and immediately following Hypothermia Treatment
Hypoxic-Ischaemic Encephalopathy (HIE) is a leading cause of neonatal mortality and morbidity. Therapeutic Hypothermia (TH) is the only treatment for HIE, which has reduced the rate of mortality and morbidity although still around fifty per cent of infants affected die or have a neurodevelopmental disability, such as cerebral palsy. Every infant with HIE receives the same TH protocol: 72 hours at 33.5 degrees Celsius, before being rewarmed slowly over 12 hours. The infants with moderate-severe HIE are at the highest risk of disability or death. Should it be possible to identify those infants at greatest risk during the TH, they could be targeted for modified TH treatment or for additional adjuvant treatments such as allopurinol, melatonin or ethyropeitin. The severity of the HIE, and prognosis on outcome, is challenging while the infant is cooled, as the cooling itself reduces the prognostic power of many of the previously accepted assessment methods. MRI (Magnetic Resonance Imaging) is the gold standard, but moving infants to an MRI whilst cooled is problematic, and the HIE injury to the brain may not be evident for several days. The objective of this research study was to find a clinical parameter which would identify those infants at highest risk of a poor outcome. HIE is characterised by hyper-perfusion, cerebral blood flow (CBF) exceeding metabolic demand, consequent from a physiological response preferentially redirecting blood flow to the brain during a hypoxic episode. Thus, a marker of cerebral perfusion that could classify HIE infants by the degree of hyper-perfusion was sought. Initially the Resisitive Index (RI), as determined by measuring the Cerebral Blood Flow Velocities (CBFV’s) using Doppler Ultrasound (DUS), was investigated as a possible cerebral perfusion marker. The RI is an accepted measure of the resistance of the cerebral arteries and an RI < 0.55 is considered to be indicative of a previous hypoxic event. In the first part of this study, the RI was evaluated in 80 healthy and 18 HIE infants and found to be unreliable; some infants with HIE and a poor outcome did not have an RI < 0.55, whereas some infants without HIE did. Moreover, DUS is not continuous; it is a one-off measurement requiring a skilled operator. A pursuit for another cerebral perfusion marker ensued. Frequency-Domain Near Infrared Spectroscopy (FD-NIRS) promised absolute measurements of oxy- and deoxy-Haemoglobin (HbO and HbR) in the brain tissue. Increased CBF was expected to result in a higher than normal concentration of blood and therefore Haemoglobin in the brain tissue. The results from the second part of this research study of 40 healthy and 6 HIE infants suggest that hyper-perfusion is expressed in the concentration of HbO in the brain tissue, and that this is most related to poor outcome. The limited HIE population in this study prevents generalisation from this finding, though suggests that there is justification for further research using FD-NIRS to monitor infants with HIE.
Dynamic stability and variability of perturbed walking in young adults
Falls are the third major cause of inadvertent injury in young Australian adults aged between 18-35 years. The inability of an individual to respond to external perturbations due to walking on an inclined surface, or internal perturbations such as dual-task walking, are known to be associated with significantly higher risk of balance loss. Significant factors known to increase risk of balance loss during walking include performing an additional task requiring high motor-cognitive, sensory or cognitive load (internal perturbations), and walking on uneven surfaces such as sloped terrain (external perturbations). At present, however, dynamic balance of the entire body and the risk of balance loss during walking under such perturbations is not well understood. The objective of this study was to investigate dynamic stability and variability of the human body during walking, and assess the influence of external, motor-cognitive, sensory and cognitive perturbations on dynamic balance, including surface inclination, use of a cell phone, auditory and visual stimulation, and mental calculation. Nineteen healthy young adult males were recruited. Three-dimensional joint kinematics were obtained using an optical motion capturing system as subjects walked at their self-selected speed on an instrumented treadmill. Dual-tasking was simulated by subjecting participants to motor-cognitive, visual, cognitive and auditory perturbations during walking including cell phone usage (talking, texting and reading), watching a video clip, listening to music, and performing numeric calculations mentally. External perturbations were also applied through alteration of surface inclination. Variability analysis was performed on spatiotemporal gait parameters using Detrended Fluctuation Analysis (DFA) and Standard Deviation. Dynamic stability was subsequently estimated for the entire body as well as the head, trunk and lower extremity joints using linear and nonlinear measures including Margin of Stability (MoS), Lyapunov Exponent (LyE) and Maximum Floquet Multipliers (MaxFM). A novel method was devised to assess stability using Margin of Stability at heel contact (HC) and minimum foot clearance (MFC), gait events associated with backward and forward balance loss, respectively. Slip and trip propensity estimated using Required Coefficient of Friction (RCoF) and MFC height, respectively. Finally, the most destabilizing additional task while walking was determined using deviation of MoS and trip propensity values during dual-task trials from the corresponding values during baseline walking. The results showed that dual-tasking during walking adversely affects balance in a direction specific-manner. Specifically, cell phone texting and reading while walking reduces balance in the mediolateral direction, while cell phone talking increases the risk of tripping in the anteroposterior direction. Upslope terrain increased the risk of balance loss in the anteroposterior and vertical directions and did not affect gait balance in the mediolateral direction, while walking down was associated with greater stability in the anteroposterior direction. Cognitive and sensory perturbations affected gait balance mostly in the anteroposterior and vertical directions rather than the mediolateral direction. Analysis of trip propensity showed that motor-cognitive dual-tasking due to cell phone usage while walking, cognitive and sensory perturbations due to performing additional auditory and visual tasks while walking are associated with greater risk of tripping, as measured by a lower MFC height. Particularly, talking while walking, and cognitive and sensory dual-tasking while walking may ultimately lead to an increase in risk of tripping in young adults. However, the risk of tripping in young individuals is not sensitive to external perturbations caused by sloped terrains. Participants mostly changed their step length and step time during walking under perturbations. Among the various measures used to determine the most destabilizing secondary task while walking, MFC height was more sensitive to the applied perturbations. Talking while walking was associated with the largest deviation from baseline condition. The findings of this investigation confirmed that head stabilization during ambulation has higher priority compared to other segments, and individuals try to adopt different strategies to attenuate perturbations from the lower body to the head. The current data highlighted the importance of arm swing in balance maintenance during walking under perturbations, and demonstrated that individuals try to compensate restricted arm swing during walking by modulating step width. With respect to gait adaptations, the results of this research support the idea that individual’s response to applied perturbations through dual-tasking while walking depend on the magnitude of the applied perturbation. The evidence from this study suggests that talking while walking is the most challenging secondary task during locomotion among the applied perturbations in this study, and additional sensory tasks are the least challenging one. These findings have significant implications for development of a gait training protocol for more frail people to successfully address common perturbations arising from daily living activities during everyday life. These data also suggest that MFC height analysis and local stability analysis of the lower body should be performed to gain better understanding on the effect of additional concurrent task while walking on the risk of tripping and gait stability, respectively. The analysis of MoS presented extends knowledge of step-to-step balance changes during walking at different gait events associated with common fall patterns occurring at HC and MFC. Further work needs to be done to analyse the influence of similar attention-demanding secondary tasks or ‘distractions’ in more vulnerable populations, including the elderly, fallers, and individuals with sensory or motor impairments that affect locomotor control.
Localisation of the Epileptogenic Zone from Interictal State MEG Data of Focal Epilepsy Patients
Over twenty million people in the world have drug refractory epilepsy. Their seizures cannot be adequately controlled by medication. Epilepsy surgery can remove or alter abnormal brain areas where seizures start, which is the only way to cure epilepsy and can be an effective treatment for drug refractory epilepsy patients. Accurate localisation of the epileptogenic zone (EZ) is crucially important to achieve seizure freedom after surgery. Magnetoencephalography (MEG) is a non-invasive brain functional imaging technique with superb temporal resolution. Clinically, neurophysiologists visually annotate interictal spikes in MEG recordings and apply localisation methods using averaged spikes. The manual annotation of spikes can be very time consuming for neurophysiologists. The aim of this thesis is to develop automatic methods to localise the epileptogenic zone using interictal state MEG recording for focal epilepsy patients. This thesis comprises three research objectives to achieve this goal. First, investigate localisation performance of kurtosis beamforming with various source selection approaches; the use of a 1 second sliding window for calculating kurtosis is shown to deliver the best performance relative to the other measures considered. Second, develop a method to detect interictal spikes automatically on virtual sensors (VSs) that are reconstructed using beamforming; the method based on feature extraction and machine learning delivers similar performance to labels by expert raters. Third, explore spatial distribution of interictal spikes across VSs and associate spike frequencies with EZs and surgical outcomes; the method is promising for localising the EZ at the lobe level. Potential future research is discussed based on these outcomes.
Development of next generation biodegradable drug-eluting coronary artery stents
Cardiovascular diseases are a leading cause of mortality globally, causing approximately 17 million deaths annually. Additionally, this number is predicted to rise to 23 million by 2030. The most common type of cardiovascular disease is coronary heart disease, a disease of coronary arteries that supply oxygen rich blood to the heart. Coronary artery disease is the build-up of a waxy substance called a plaque inside the coronary artery which leads to its narrowing and blockage. In current medical practice, coronary heart disease is commonly treated through balloon angioplasty and stenting to open the artery. Current stents are drug- eluting, metallic, and permanent, and recipients require prolonged anti-platelet therapy. Permanent stenting is not required. The diseased vessel can heal within 6 months to 1 year after intervention. As such, the concept of biodegradable stents has emerged as the alternative to conventional stenting, in which the stent degrades away leaving behind only the healed vessel. The first generation of biodegradable stents has been linked to higher rates of late stage thrombosis, and it has been suggested that this is due to increased strut thicknesses that cause disturbance to the laminar blood flow and result in activation of thrombogenic pathways. The aim of this thesis is to develop customizable, biodegradable, multi drug eluting coronary artery stents by using polymer chemistry, materials science, and additive manufacturing. The novel materials developed in this work are to be blood-compatible, biodegradable, have sufficient mechanical properties, promote endothelialisation, have multi-drug eluting properties, and be processable through additive manufacturing techniques. To achieve this, we used the following approaches: (1) Design and additive manufacturing of custom-made biodegradable nanocomposite based coronary artery stents (2) Design and synthesis of biocompatible and biodegradable core-cross linked star-brush polymers for antithrombotic drugs (3) Development of multi-drug eluting biodegradable nanocomposite-star polymer materials for application as coronary artery stents utilizing additive manufacturing.
Selenium nanoparticles as antibacterial agents for potential application in chronic wound healing
Chronic wounds have become a global problem. The importance of microbial colonies in delaying chronic wound healing has been highlighted recently. Keeping the wound free of infection plays a vital role in fast and successful wound healing. The rapid development of antibiotic resistance and the inability of antibiotics to penetrate biofilms seriously limit the efficacy of antibiotics. Therefore, new effective antimicrobial treatments are urgently needed to realize rapid and successful chronic wound healing. One promising candidate to address this requirement is selenium nanoparticles (Se NPs) which have antimicrobial activity. In this thesis, we were working on developing new antibacterial Se NPs, and adopting 3D printing technology to fabricate latticed wound dressings with a controlled release of Se NPs. First, the influence of size on the antibacterial activity and cytotoxicity of Se NPs was investigated. In Chapter 3, spherical Se NPs ranging from 43 to 205 nm in diameter were fabricated, and their mammalian cytotoxicity and antibacterial activity as a function of their size were systematically studied. The antibacterial activity of the Se NPs was shown to be strongly size dependent, with 81 nm Se NPs showing the maximal growth inhibition and killing effect of methicillin-sensitive and methicillin-resistant Staphylococcus aureus (MSSA and MRSA). The Se NPs were shown to have multi-modal mechanisms of action that depended on their size, including depleting internal adenosine triphosphate (ATP), inducing reactive oxygen species (ROS) production, and disrupting membrane potential. All the Se NPs were non-toxic towards mammalian cells up to 25 microgram/mL. Furthermore, the minimum inhibitory concentration (MIC) for the 81 nm particles produced in this research against MSSA is 16 +/- 7 microgram/mL, significantly lower than previously reported MIC values for Se NPs. This data illustrates that Se NP size is a facile yet critical and previously underappreciated parameter that can be tailored for maximal antimicrobial efficacy. We have identified that using Se NPs with a size of 81 nm and concentration of 10 microgram/mL shows promise as a safe and efficient way to kill S. aureus without damaging mammalian cells. Second, the effect of charge on the antibacterial activity of Se NPs was researched. It has been shown previously that Se NPs with a net negative surface charge have good antibacterial activity against Gram-positive bacteria but are less effective against Gram-negative bacteria. Gram-negative bacteria have been observed to be more sensitive to positively charged nanoparticles because the surface charge of Gram-negative bacteria is generally more negative than that of Gram-positive bacteria. Therefore, in Chapter 4, Se NPs were capped with a positively charged protein – recombinant spider silk protein eADF4 (kappa 16) – to give them a net positive charge. Compared to the negatively charged polyvinyl alcohol (PVA) capped Se NPs, the positively charged eADF4 (kappa 16) coated Se NPs demonstrated a much higher bactericidal efficacy against the Gram-negative bacteria E. coli in water. Particularly, the minimum bactericidal concentration (MBC) of 46 nm eADF4 (kappa 16) capped Se NPs (8 +/- 1 microgram/mL) was approximately 50 times lower than the 46 nm PVA capped Se NPs (405 +/- 80 microgram/mL). Scanning electron microscopy (SEM) images showed that the PVA capped Se NPs were repelled by the E. coli cells, while the eADF4 (kappa 16) capped Se NPs attached to or even coated the E. coli cells. In addition, antibacterial films were created by immobilising the eADF4 (kappa 16) capped Se NPs on positively charged spider silk and these were shown to retain good bactericidal efficacy and overcome the issue of low particle stability in culture broth. It was found that these Se NPs needed to be released from the film surface in order to exert their antibacterial effects and this can be achieved by regulating the surface charge of the film. Overall, eADF4 (kappa 16) coated Se NPs are promising new antibacterial agents against life-threatening bacteria. Third, the optimal size of Se NPs combined with positive surface charge was fabricated to realize a high antibacterial efficacy. In Chapter 5, 81 nm Se NPs combined with epsilon-poly-L-lysine (Se NP-epsilon-PL) were fabricated and their antibacterial activity and cytotoxicity were investigated. Se NP-epsilon-PL exhibited effective antibacterial activities against all tested 8 different species of bacteria including both Gram-positive and Gram-negative bacteria, and some of them are drug-resistant bacteria types. Bacteria were found to be very difficult to develop resistance to Se NP-epsilon-PL comparing to the conventional antibiotic kanamycin. S. aureus and E. coli started to develop resistance to kanamycin from 44 and 52 generations, respectively. By contrast, S. aureus started to develop resistance to Se NP-epsilon-PL after 132 generations, and E. coli failed to develop resistance to Se NP-epsilon-PL during the whole tested 312 generations. On the other hand, Se NP-epsilon-PL showed low cytotoxicity to human dermal fibroblasts. Based on its efficient and wide-spectrum antibacterial activity, the difficulty to develop resistance in bacteria and its low cytotoxicity, Se NP-epsilon-PL might become a promising member of the new generation of antibacterial agents. Finally, in Chapter 6, alginate dressings with pH-responsive release of Se NPs were fabricated using 3D printing technology. Although the antibacterial activity of Se NPs has been proved, relatively high concentrations of Se NPs are toxic to the mammalian cells. It has been reported that the pH of body fluid in bacteria infected wounds is higher than that in normal skin. To enable faster release of Se NPs at a relatively high pH to perform higher bactericidal efficacy and slower release of Se NPs at a relatively low pH to protect the normal cells, alginate wound dressings with pH-responsive release of Se NPs were 3D printed using a Bioplotter. Calcium phosphate nanoparticles (CaP NPs) were introduced into alginate dressings to make their degradation pH-responsive, resulting in pH-responsive release of Se NPs from these dressings. The dressings’ mechanical properties, degradation rates and releasing rates of Se NPs were investigated. The results showed that the addition of CaP NPs can increase both tensile strength and elongation of alginate dressings and make the degradation rate of alginate dressings faster at a relatively high pH than that at a relatively low pH. Similar to these degradation results, when Se NPs have been introduced, the release rate of Se NPs from the scaffolds at a relatively high pH also showed faster than that at a relatively low pH. In conclusion, the potential of using Se NPs as antibacterial agents has been investigated, and 3D printed alginate dressings with pH-responsive release of Se NPs have been developed.
An EMG-driven neuromusculoskeletal modelling framework for evaluation of shoulder muscle and joint function
Knowledge of shoulder muscle and joint forces during daily living activities is important for developing injury prevention strategies, tailoring exercise therapies and rehabilitation, and design of joint replacements. As non-invasive measurement of muscle force is currently not possible, computational strategies have been widely employed to solve this challenge. The muscle force equilibrium problem at the shoulder complex is statically indeterminate, due to the higher number of joint actuators than degrees of freedom of joint movement. This precludes a unique solution of muscle forces. Optimisation methods, which are most commonly used to solve this indeterminacy problem, have been used to decompose calculations of net joint moments into muscle forces; however, it is thought that these approaches underestimate muscle co-contraction. The objective of this thesis was to develop the first fully electromyography (EMG)-driven neuromusculoskeletal modelling framework to evaluate shoulder muscle and joint function. The specific aims were to use this framework to (i) evaluate neuromusculoskeletal function of the shoulder during upper limb activities of daily living, including muscle and joint forces (ii) validate model function by comparing muscle-generated joint moments with inverse dynamics joint moments and comparing glenohumeral joint force calculations with in vivo measurements using instrumented implant data (iii) investigate the effect of EMG-driven neuromusculoskeletal model calibration strategies on muscle and joint force calculations, and (iv) compare muscle and joint force predictions from the EMG-driven neuromusculoskeletal model with calculations using static optimisation. Four healthy adults performed three groups of functional upper limb tasks representing activities of daily living: active abduction, flexion, reaching and head touching, mid-range submaximal isometric abduction, adduction, flexion, extension, internal rotation and external Abstract ii rotation, as well as passive functional abduction and flexion. EMG data from 16 shoulder muscles were simultaneously acquired using surface and intramuscular electrodes, and upper limb kinematic and external forces measured. Subject calibrated neuromusculoskeletal models were developed for each subject using OpenSim. Muscle forces, resultant joint torques, and contact forces were subsequently calculated using EMG data, joint kinematics, and external forces as inputs to the model. Anterior and middle deltoid were the major arm elevators and generated the greatest muscle force among all glenohumeral muscles. Infraspinatus and supraspinatus were active in all upper limb tasks investigated and were also important contributors to joint motion and stability. Glenohumeral joint forces showed high agreement with previously reported in vivo instrumented implant data, while muscle-generated joint moments showed close agreement with net joint moments calculated using inverse dynamics. Calibration of the EMG-driven neuromusculoskeletal models over a wide range of contractions performed in multiple movement planes led to more globally representative musculotendon parameters that produced more accurate net joint moment and joint force predictions. Calibrating models using tasks in one plane, even when using the calibrated model to predict muscle and joint forces in that plane, produced joint force estimates that are less accurate than when models were calibrated over multiple planes. Simulation of submaximal isometric contractions showed that the EMG-driven model predicted antagonistic muscle function for pectoralis major, latissimus dorsi and teres major during abduction and flexion; supraspinatus during adduction; middle deltoid during extension; and subscapularis, pectoralis major and latissimus dorsi during external rotation. In contrast, static optimisation neural solutions showed little or no recruitment of these muscles, and preferentially activated agonistic prime movers with large moment arms. As a consequence, glenohumeral joint force calculations varied substantially between models. This thesis presents the first shoulder muscle and joint force solutions derived from a neuromusculoskeletal model driven entirely from EMG data measured from all the major muscles spanning the glenohumeral joint. The findings highlight limitations in current static optimisation techniques for evaluating shoulder muscle and joint behaviour and provide guidelines for EMG-driven model development and calibration. The results provide insight into the role of the muscles in stabilising the glenohumeral joint during active joint motion. Because the modelling approaches developed incorporate subject-specific measures of muscle activity, they lend themselves well to future applications in evaluating upper limb muscle and joint function in cases of neuromuscular impairment, including stroke, traumatic brain injury, and spinal cord injury.
Assessment of a pressure-cast socket for transtibial prostheses in under-resourced environments
Current socket casting methods used in under-resourced environments are highly reliant on the skill and experience of the prosthetist. Pressure-casting methods have been advocated to reduce skill dependency in socket casting. To date, however, socket production studies have only been performed in resourced settings and involved qualified prosthetists. In this thesis, a pressure-casting method to reduce the skill dependency in socket casting has been described in detail. Appropriate and feasible measures to assess wearer outcomes comprised standard functional tests, spatio-temporal gait parameters, subjective measures of user satisfaction and quantification of the limb-socket interface pressures. The participants of all studies herein were adults with unilateral transtibial amputations. All studies were conducted in Vietnam and utilised pressure-cast (PCAST) sockets with low-cost polypropylene components. All sockets were cast by local orthopaedic technicians when appropriate. Wearer outcomes with the PCAST sockets were similar to the participants’ original prosthetic limbs and high levels of satisfaction were recorded following an extended usage period of five months. Temporal gait changes suggested greater willingness to load the prosthetic limb over the usage period. In contrast, satisfaction and comfort data suggested that feelings of comfort and stability decreased over the usage period. As a baseline, the participants’ original prostheses were not consistent due to varying age, quality and socket design. Thus, the outcomes of participants fit with both PCAST and PTB sockets were compared. The results showed no significant differences in the initial participant functionality, spatio-temporal gait characteristics, gait symmetry, or comfort between the skill-dependent PTB socket and the PCAST socket. PCAST socket fit was investigated by examining interface pressure distribution, magnitude and duration and exploring how these pressures vary with wearer comfort. The pressure distribution was non-uniform with high pressures identified at the bony prominences, especially the tibial crest. The duration, rather than magnitude, of these pressures, appeared to influence wearer comfort. High pressures in the anterior proximal region and longer loading durations at the lateral proximal and medial distal regions also potentially affected wearer comfort. These data may assist future researchers in analysing and interpreting socket interface pressures. The staged phasing of data collection periods permitted continual development of the PCAST technique. Wearing a thick sock during casting appeared to reduce tightness and pain around the tibial condyles. Thus, the use of a thick cotton sock during casting was deemed beneficial. Further research, including interface pressure analysis, is required to determine if using a Pelite liner with the PCAST socket improves wearer outcomes. This research has shown the PCAST technique is able to produce a functional, biomechanically sound and satisfactory socket for prosthetic users in under-resourced environments. Additional studies have been recommended based on the outcomes and limitations of the studies herein to further the understanding and potential for success for the PCAST socket in under-resourced environments.
Acoustically and electrically evoked cortical potentials for hearing threshold estimation
Hearing thresholds are frequently used in clinical audiology for hearing assessment or cochlear implant programming. Hearing thresholds are difficult to determine in non-responders such as infants. However, the accuracy of threshold estimates is important because they allow diagnosis of hearing loss as well as optimal implant programming, which in turn is important for ensuring good language development and speech understanding performance. The aim of this research was to develop an objective and clinically relevant method to estimate hearing thresholds. Current objective measures are either not accurate enough to be used in a clinical setting, or have prohibitive equipment or testing time requirements, barring their adaptation as a clinically used tool. Therefore, the research methodology of this work was guided by clinical applicability. Cortical auditory responses to acoustic and electric stimuli were collected from 20 otologically-normal adults and from 20 adults with cochlear implants. The scope of the research was restricted to analysis techniques that relied on single-channel electroencephalography (EEG) measurements in order to minimise requirements for equipment and test set-up time in any resulting clinical application. The first study was an exploration of the EEG features that could potentially be used for threshold estimation. The objective was to find EEG features that varied as the stimulus level varied. In this study I presented sounds at a variety of intensities and examined a selection of EEG features evoked by the stimuli. I evaluated the EEG features by comparing how a classifier using each feature was able to distinguish between EEG responses to the stimuli at different levels. I found that the phase-locking value was the optimal feature when distinguishing between EEG responses to sounds of different intensities. In addition, I found that the EEG dynamics in response to each stimulus did not contain any detectable, non-phase-locked spectral perturbations. The objective of the second study was to show how these features could be used to estimate hearing thresholds. I calculated growth functions for each EEG feature as a function of stimulus intensity, and then used fits for these functions to estimate thresholds. I then validated the threshold estimates from the fitted functions against behavioural threshold estimates in the normal hearing listeners. I found that the phase-locked information in the EEG, expressed by the peak-to-peak amplitude and phase-locking value features, produced hearing thresholds that were closed to the behavioural thresholds. Threshold estimates calculated from the EEG features differed from behavioural thresholds by within +/-20 dB (95% confidence interval) across 20 subjects with 12.7 minutes testing time per stimulus. In the third study, I extended the growth function methods from the second study to cochlear implant users. After ten minutes of testing, the resulting correlation between threshold estimates and behavioural threshold was 0.92 for 20 subjects, and the accuracy was +/-22% dynamic range (95% confidence interval). This study was used to develop the method and, as such, sub- and supra-threshold stimulus levels were determined a priori from the cochlear implant users’ clinical maps. However, this would be difficult in a clinical setting, where there may be no information about the hearing of the patient. Thus, in the final study, an adaptive algorithm to automatically determine stimulation intensities was developed, simulating real clinical use. The algorithm could estimate thresholds as accurately as previous studies which used pre-determined stimulus intensities but only if it was given additional time, which may be prohibitive in a clinical setting. Suggestions for further adaptive algorithm improvements are also presented. This thesis presents a clinically relevant and fully automated method to estimate hearing thresholds using the cortical EEG response, and paves the way for hearing devices or clinical devices to incorporate this novel technology.
Learning receptive field properties in a biologically plausible model of primary visual cortex
While the eye is a complex structure, it is just the start of an even more complex series of visual information processing centers in the brain. Around 40% of the human cortex is involved in vision processing. The areas of the brain that process visual information are divided into distinct compartments. The primary visual cortex, area V1, is the first stage of visual processing in the cortex. In general, the visual cortex processes visual information in a hierarchical structure: from thalamus to V1, V1 to V2, and V2 to higher levels. However, many of the connections do not follow these strict hierarchical rules and interconnections between brain areas are very common. In V1, simple and complex cells are two distinct categories of cells. However, how the properties of receptive fields (RFs) for simple and complex cells are learned still remain unclear. Artificial neural networks are bio-inspired and powerful tools for tasks such as image recognition. Some of the networks can even generate RFs that are very similar to the simple cells in V1. However, there are currently many aspects of neural network models of biological systems that are biologically implausible. We might understand biological vision processing better by incorporating biological constraints in artificial neural networks. This thesis focuses on V1 and builds biologically plausible models for simple and complex cells by incorporating biological constraints in artificial neural networks. Our results demonstrate that a two-layer model of the visual pathway from the lateral geniculate nucleus to V1 that incorporates biological constraints and efficient coding can account for the emergence of many experimental phenomena of simple cells when the model is trained on natural images. The model demonstrates that efficient coding can be implemented by the V1 simple cells using neural circuits with a simple biologically plausible architecture. Our model of complex cells, based on the Bienenstock, Cooper and Munro (BCM) rule, demonstrates that properties of RFs of complex cells can be learned using a biologically plausible learning rule. Quantitative comparisons between the model and experimental data are performed. Results show that model complex cells can account for the diversity of complex cells found in experimental studies. These findings help us to better understand biological vision processing and provide us with insights into the general signal processing principles that the visual cortex employs to process visual information.
A bond graph approach to integrative biophysical modelling
A major goal of systems biology is to develop comprehensive, multi-scale mathematical models of physiological systems that integrate biological knowledge from the scale of molecules to the scale of tissues and organs. Models on this scale hold great potential in advancing our knowledge of biology and medicine, but they have yet to be achieved in complex biological systems. It is widely acknowledged that constructing large-scale models requires the reuse and integration of existing models; however, model integration is currently challenging because many existing models violate the conservation laws of physics, especially conservation of energy. It is therefore highly desirable to express models in a framework that respects the laws of physics and thermodynamics. Bond graphs are an energy-based modelling framework, initially developed for use in multi-physics engineering systems to help derive equations consistent with the laws of physics. More recently, bond graphs have been applied to the field of biology where they have helped in making models physically and thermodynamically consistent. While bond graphs provide several advantages for large-scale modelling such as thermodynamic consistency and hierarchical modelling, they have yet to be applied to large-scale dynamic models of biological systems. This thesis aims to develop methods based on the bond graph framework to facilitate model reuse and integration. These methods are demonstrated by applying them to biomolecular systems within the cardiac cell. Firstly, bond graphs are applied to membrane transporters, demonstrating that bond graphs can be used to correct thermodynamic inconsistencies within existing models. Secondly, independently developed models of ion channels and transporters are coupled into a model of cardiac electrophysiology, showing that bond graphs can be used to systematically explain the issues of drift and non-unique steady states that affect many existing models. Finally, a generalised method for simplifying models of enzyme kinetics is developed and used to facilitate the development of simple, thermodynamically consistent models of enzymes that are easily incorporated into larger models.
Dynamics of functional brain connectivity in schizophrenia: machine learning models for diagnosis and prognosis
To date, most studies of resting-state functional connectivity implicitly assume that connections remain unchanged over time. However, recent studies suggest that functional connectivity across a range of species exhibits time-varying behaviour. Understanding the time-varying properties of functional connectivity appears particularly beneficial in studying disorders such as schizophrenia. This thesis scrutinises the potential applications of dynamic functional connectivity (dFC) in aiding both the diagnosis and prognosis of schizophrenia. While previous dFC studies focussed on assessing variability of connection strengths over time, we propose a model that can simultaneously capture dynamics in both time and space. Temporal sliding windows were used to map dynamics over time. Spatial variability was accounted by a modified seed-based connectivity approach that allowed different network regions to vary their spatial layout, by expanding or contracting over time according to their connectivity profile. Connectivity measures based on the proposed method were then compared to those derived from traditional static and temporally dynamic connectivity, in predicting the diagnostic status in schizophrenia using support vector machine-based classifier models. Prediction accuracies exceeding 91% were obtained with our method, while previous methods yielded significantly lower accuracies. This suggests that the proposed method provides a better characterisation of connectivity dynamics and extracts novel disease-specific information that can potentially yield new insights into the pathophysiology of schizophrenia. Compared to healthy individuals, schizophrenia patients exhibited both temporally and spatially diminished, but more variable functional connectivity across different resting-state networks. Further, dynamic interactions among different resting-state networks were characterised using a hidden Markov model (HMM). Fluctuations in fMRI activity within 14 canonical networks, derived from both healthy individuals and schizophrenia patients, were concatenated and then quantized into 12 states using the HMM. We observed that patients spent significantly greater amounts of time in states characterised by low default-mode network (DMN) activation and heightened activity within different sensory networks. It was also found that patients lacked the ability to effectively up/downregulate the activity within the DMN. Furthermore, measures of dynamics derived from the model associated significantly with positive symptoms of schizophrenia and provided high predictive diagnostic accuracy (~85%). Finally, we examined the prognostic predictive power of dFC measures. Specifically, we tested if measures derived from dynamic connectivity among the DMN regions aid in classifying patients into worsening or improving in symptom severity after a year. Classifiers trained on DMN connectivity dynamics yielded 75-80% accuracies in predicting prognostic status in all the three types of scores considered (positive, negative and overall symptom severity). Importantly, dynamic connectivity measures were found to be better predictors than other, previously proposed variables such as cortical thickness, grey matter volume, clinical and behavioural measures and static connectivity. Together, the analyses presented in this thesis validate the utility of dynamic functional connectivity in characterising schizophrenia pathology and in aiding the adoption of more evidence-based treatment options.