Biomedical Engineering - Theses
Permanent URI for this collection
Search Results
1 - 10 of 77
-
ItemInvestigation of Speech Imagery for Brain-Computer Interfaces( 2023-06)Brain-computer interfaces can restore various forms of communication in paralyzed people who have lost their ability to articulate intelligible speech. These devices, known as speech neuroprostheses, serve as direct interfaces between living brains and artificial electronic components. The field of research is rapidly moving toward clinical trials with potential users benefiting from chronic brain implants. However, research with non-target participants is essential to accelerate the development of such devices. This thesis investigated the use of speech imagery tasks in study participants implanted with intracranial electrodes to improve the performance of speech neuroprostheses. Four original studies were conducted to address practical considerations related to: (1) the identification of discriminative features for keyword detection from brain recordings, (2) the characterization of brain activation patterns during the production of isolated speech sounds, (3) the implementation of a closed-loop algorithm for real-time speech synthesis from brain recordings, and (4) the importance of brain coverage and task instructions to synthesize intelligible artificial sounds during silent speech. In all four studies, patients with medication-resistant epilepsy were temporarily implanted with intracranial electrodes, either stereotactic electroencephalography (SEEG) depth or electrocorticography (ECoG) surface electrode arrays, for the localization of the seizure onset zone prior to brain resection. They were asked to perform overt and silent speech tasks while their brain signals were recorded. The first study adopted a traditional decoding approach based on discrete trial classification and found that high-gamma activation in the superior temporal gyrus (STG) was the most discriminative feature for keyword detection during overt speech. No discriminative feature was found during imagined speech, which highlighted the need to go beyond trial-based designs. The second study introduced a voice-based cursor control task through the production of isolated phonemes. Onset and sustained neural activation patterns were observed in the STG but trajectory reconstructions from intracranial signals remained low. The mismatch between the control task and visual feedback also constituted an important challenge for restoring intuitive communication. The third study characterized the trade-off between decoding performance and execution times in the proposed algorithm for real-time speech synthesis from intracranial recordings. Purely based on acoustic features of speech, the model could be trained with a small patient-specific dataset and immediately tested in any language with no assumption on brain coverage. The fourth study tested the proposed closed-loop system in ten participants with various coverage of brain areas. Artificial sounds were synthesized from the STG during overt speech in three participants and from the precentral gyrus during mimed speech in two participants. Human perceptual judgments supported the fact that some of these sounds were occasionally intelligible. Unfortunately, no artificial sounds were synthesized during imagined speech. These four studies made specific contributions to the field of research by going beyond trial-based design, understanding the gap between voice and brain control, synthesizing artificial sounds in real time under clinical constraints, and rethinking the utility of mimed and imagined speech in able-bodied study participants. Altogether these four studies added further evidence toward the feasibility of closed-loop speech neuroprostheses that continuously synthesize intelligible artificial sounds. Such brain-to-audio systems have the potential to restore functional communication at a conversational rate of 150 words per minute, outperforming state-of-the-art brain-to-text systems that operate at 62 words per minute. This thesis concludes with a thorough discussion on the limitations of the proposed brain-to-audio approach and future directions to overcome the remaining barriers to clinical translation in target patients.
-
ItemDevelopment and Validation of Image Processing Framework for Dynamic Quantitative Morphometric Analyses of Joint and Osteochondral Tissues( 2023-03)Quantitative morphometric analysis (QMA) using micro-computed tomography (microCT) allows non-invasive assessment of structural degradation of joint tissues, such as bone and cartilage, during osteoarthritis (OA) progression in small animal knees. Whole-joint features, such as joint space width and joint alignment, were shown to provide sensitive markers of biomechanical changes due to OA. As elastic links, the relative position of joint components can vary. Combined with traditional manual image processing and analysis approaches that requires recalibration between studies, QMA of the joint often encounters reproducibility issues that are technically challenging and time-consuming to tackle. The general aim of this thesis is to examine how improvements in image processing protocols can lead to reproducible and dynamic quantitative analysis of joint morphology. In this work, reproducibility issues in the acquisition, processing, and analysis routines were tackled by addressing the following hypotheses: 1. Introduction of contrast agents and control of knee pose with a positioning device during microCT acquisition allows reproducible and accurate in situ QMA of cartilage and joint. 2. Automating the image processing pipeline through morphological representation using spherical harmonics improves the efficiency and reproducibility of joint QMA. 3. An empirical probabilistic approach provides an automatic and model-invariant solution to statistically capture morphological changes due to OA by detecting osteophyte activity. In response to the first hypothesis, a whole animal positioning device and a cationic contrast agent injection protocol were developed for in situ microCT morphological assessments of the mouse knee. By securing the mouse using the device for consistent landmark location and limb stabilisation, high reproducibility of joint QMA results was achieved. Combined with the optimised contrast agent injection protocol, accurate measurements of cartilage attenuation and QMA were concurrently observed. Addressing the second hypothesis, an automatic workflow based on spherical harmonics and persistent homology was developed to process the acquired microCT images for joint QMA. Spherical harmonics, describing the basic shape of the tibia, were used to align the joint to a common reference coordinate system. Anatomically meaningful subdivision of the joint into lateral and medial volume of interests was subsequently performed using a watershed method based on persistent homology. Validated on microCT scans of rabbit and rat knees, joint QMA results from images processed through the automated workflow show excellent reproducibility with significantly reduced time and personnel training requirements compared to manual processing. Tackling the final hypothesis, an empirical probabilistic approach was developed for an automatic and model-invariant QMA of osteophyte activity. Osteophyte disrupts the integrity of the cortical exterior, leading to a rougher surface with smaller local thickness values. The method statistically captures these changes by identifying shifts in local thickness as an indicator of osteophyte activity. Validated on microCT datasets of rabbit and rat knees, the method and the resulting models were shown to be objective and invariant to animal scale. Using this approach, reproducibility can be maintained across studies by removing the need for parametric recalibrations to adjust for animal scale and voxel size. As these approaches are invariant to animal scale and voxel size, the need for recalibration in subsequent studies has been removed. By improving the acquisition, processing, and analysis protocols, this work increases the reproducibility and efficiency of QMA workflows, leading to higher throughput and more dynamic QMA outcomes.
-
ItemOptimizing Acoustic Systems for Biomedical Applications with Numerical Modeling( 2023-06)This thesis explores the development and optimization of acoustic and acoustofluidic devices. Acoustofluidic devices, which combine principles of acoustics and microfluidics, have emerged as a promising platform for biological micro-object micromanipulation due to their non-invasive, accurate, rapid, and label-free qualities. Acoustofluidic devices have found utility in various biomedical applications including single-cell studies, point-of-care testing, lab-on-a-chip studies, and tissue engineering. The objective of this thesis is to develop key components of these devices and explore new device configurations employing computational modeling and optimization techniques. As a result, novel device configurations are developed enabling complex and high-resolution micromanipulation of suspended micro-objects. In the studies presented here, computational analysis is utilized to optimize (1) traveling surface acoustic wave device dimensions, (2) the configuration of a planar acoustic resonator that integrates a structured surface, (3) the thickness of the coupling layer and superstrate materials for bulk-wave transmission, (4) the shape of acoustically-actuated 3D microstructures, and (5) waveguide topology in reusable face masks. In doing so, this work demonstrates that computational analysis is an integral part of the development of acoustofluidic devices for advanced micromanipulation and sound-transmitting structures, which have extensive potential in biomedical applications.
-
ItemMechanoregulatory Communication and Remodelling Networks in Cartilage Tissue Models( 2023-05)The structural integrity of cartilage is vital for healthy and pain free ageing. However, it is a slow remodelling tissue and has limited healing capacity. To successfully replicate native cartilage tissue, it is crucial to understand its structure-function relationship and mechanical complexity, and to have platforms to study subsequent cellular remodelling behaviour. Therefore, this work aims to understand cartilage micromechanics by using an image-guided micromechanical evaluation approach and developing a cartilage tissue model to mimic heterogeneous mechanical domains.
-
ItemMagnetic manipulation of cells to enhance tissue engineering( 2023-03)One of the major detrimental effects of the aging process is the natural atrophy of skeletal muscle tissues, a process where individuals typically experience lower muscle mass, reduced muscle function and, as the problem worsens, compromised personal independence. This problem can become exacerbated due to injuries and diseases such as cancer, where patients may suffer from cachexia, a severe form of atrophy which leads to a 20-70% total volume loss in affected muscles. As it stands now, in vitro tissue engineered muscle fibres are often-times functionally immature, making it difficult to use these fibres as experimental models for drug testing. Part of the problem is that current tissue engineering methods use complex and expensive bioreactor systems to exercise skeletal muscle cells in vitro, systems that are inherently inflexible in terms of their potential applications. This technological shortcoming limits the insights that can be gained about muscle development and its disease states. Therefore, the aim of this thesis was to develop an in vitro magnetic stimulation method that provides a finer, more flexible alternative to typical bioreactors, and to investigate possible applications of this novel stimulation method for C2C12 myoblasts grown in a variety of 2D and 3D environments. To achieve this aim, magnetic chitosan microspheres (1-10 micron diameter) were produced and loaded with 30 nm, polyethylene glycol (PEG)-coated iron oxide nanoparticles (Mag30-CMs). The microspheres were produced at this size to limit cellular uptake prior to stimulation, and the microspheres were further functionalized with an RGD-containing peptide (Mag30-CM-RGDs) to enable external cellular stimulation through key integrin receptors. A magnetic stimulation chamber, optimized using finite element simulations, was 3D-printed and contained a sterile culture plate and two N45-grade neodymium magnets, allowing for the culture of magnetically-labelled C2C12 myoblasts inside a controlled and well-defined static magnetic field. After this system was developed, the same stimulation regime was applied to C2C12 myoblasts grown on a soft (8 kPa) 2D gelatin methacryloyl (GelMA) hydrogel scaffold, in order to determine the efficacy of this technique on softer substrates. Finally, the stimulation regime was then applied to C2C12 myoblasts grown in RGD-functionalized chitosan-gelatin cryogels, with macrostructures consisting of interconnected aligned pores roughly 50-250 micron wide. The maturity of the differentiated myotubes produced in the three environments was characterized based on the qPCR gene expression of key myogenic regulatory factors (MRFs), as well as morphological analysis (fusion index) done via confocal imaging. The magnetic stimulation regime method was modelled to produce piconewton-sized forces directly on the cell surface, resulting in a significantly higher fusion index as well as up-regulated gene expression of key MRFs for stimulated myotubes on hard tissue culture plastic (TCP). Stimulated myotubes had a 5-fold increase in multinucleated myotubes with 4+ nuclei over control, an important indicator of maturity. Significant increases from 2-6 fold were found in the gene expression of mature myosin heavy chain (MHC) genes MHC1, MHC2x, and MHC2a, in addition to a 2-2.3 fold increase in myogenin, a key MRF. These results indicate a promotion of myotube maturity in vitro in response to magnetic stimulation via the application of Mag30-CM-RGDs in a static magnetic field on hard TCP. In comparison, the soft GelMA hydrogels did not see the same benefits from the stimulation regime, producing no significant changes between the stimulated group and control. The 3D porous cryogels, functionalized with RGD-containing peptides, demonstrated some evidence of good cell attachment at early timepoints, but the attachment was not robust enough for the cryogels to serve as a scaffold for this kind of stimulation method, and cells did not survive long enough inside the gels to undergo differentiation. In this thesis, a robust and flexible magnetic stimulation method was produced and investigated in a novel skeletal muscle tissue engineering application. The Mag30-CM-RGDs showed evidence of maintaining cell attachment after 5 days of differentiation, and were able to promote muscle maturity in a 2D setting. This technique is not unique to skeletal muscle cells, and the methods utilized here can be adapted and applied to other types of tissues. As the materials are biocompatible, future experiments can be conducted to determine if these materials can be effective in vivo as well as in vitro.
-
ItemA Feasibility Analysis of Neuromodulation Using Endovascular Focused Ultrasound( 2023-02)Neuromodulation is an effective way to treat a variety of neurological conditions and impairment. For example, deep brain stimulation (DBS) is used clinically for management of drug-resistant Parkinson’s disease and epilepsy. Neuromodulation devices are also used to evoke sensory perception when one has sensory impairment. The most notable examples are the cochlea implant and the bionic eye, though sensations of touch can also be evoked through neuromodulation. While neuromodulatory devices may be life-transforming for certain users, they are far from ideal: electrodes must be precisely placed into neural tissue, and the injected electric current is prone to spreading through neural tissue and activating unintended brain circuits. This may induce side-effects or limit the acuity of the device. In the case of the bionic eye, this limits the maximum obtainable spatial resolution. In the case of the cochlea implant, this limits pitch perception and discrimination. Fortunately, the limitations of standard electrical neuromodulation techniques may be overcome by using a non-electrical phenomenon to evoke the neural responses. For example, ultrasound can modulate neural activity. It propagates through tissue, may be focused within a narrow region of tissue, and is unaffected by tissue conductivity. For this reason, ultrasonic neuromodulation methods have come under the spotlight. Ultrasonic energy is typically delivered to the brain from a focused transducer placed on the outside of the skull. This is termed transcranial focused ultrasound (tFUS). It has the benefit of being non-invasive; however, it is difficult to focally target a small neuronal ensemble deep within tissue. This is due to the beam absorption from the skull requiring that a low ultrasound frequency be used, which in turn elongates and broadens the beam, especially since the ultrasound source is far from the target. A new method of delivering ultrasonic energy for the purpose of neuromodulation is proposed: endovascular focused ultrasound (eFUS). A small device is placed within a blood vessel in the body, and it emits an ultrasound beam focused upon a neuronal ensemble within its general vicinity. This is more invasive than tFUS, but it has the potential to induce less tissue damage than DBS where holes are drilled into the skull and long leads are permanently inserted into the brain while the patient is awake. The safety and efficacy of this new approach to neuromodulation was assessed using computational modelling. An early prototype device, consisting of polymer-based capacitive micromachined ultrasound transducers operated as a phased array, was fabricated and bench-tested. Computational modelling suggests the presence of a therapeutic window for performing neuromodulation with an eFUS device for treating Parkinson’s disease symptoms. The prototype device emitted a focused ultrasound beam when driven by only four unique signals. The eFUS method of neuromodulation overcomes many limitations of standard electrical techniques, and it has a wide scope of applications with the potential to drive the field of neuromodulation in a new direction.
-
ItemOptimised techniques for ultra-high field functional magnetic resonance imaging( 2022)Functional Magnetic Resonance Imaging (fMRI) has become broadly used to study brain function since its invention in the late 20th century, applicable to both the study of healthy controls and patient groups. Different fMRI methodologies such as Blood Oxygenation Level Dependent (BOLD) fMRI and Arterial Spin Labelling (ASL) fMRI have been developed to non-invasively measure multiple physiological information related to neuronal activity in the brain. In the last two decades, fMRI at ultra-high field (UHF) (i.e. >= 7T) has gained an increasing amount of interest, driven by the increased signal-to-noise ratio brought by the increased field strength. However, ultra-high field fMRI suffers from challenges such as signal degradation caused by increased spatial inhomogeneity of the static magnetic field (i.e. B0) and the radio-frequency (RF) field (i.e. B1+), as well as limitations caused by increased power deposition. This thesis covers three optimised techniques for fMRI at ultra-high field by focusing on BOLD fMRI and ASL, aiming to overcome aforementioned challenges. Starting with signal formation, the Hybrid Adiabatic Pulse with asYmmetry (HAPY), a new class of optimised RF pulse for Pulsed ASL (PASL), is introduced to overcome B0 and B1+ inhomogeneity and increased power deposition. The presented technique offers robust cerebral blood flow measurement with reduced energy deposition under the effect of B0 and B1+ inhomogeneity. The second technique presented in this thesis is an optimised BOLD fMRI data acquisition protocol matched to analysis settings, termed Smoothing-Matched EPI. High spatial resolution fMRI at ultra-high field faces challenges caused by the long acquisition window with increased sensitivity to B0 inhomogeneity. This technique provides enhanced BOLD sensitivity in high spatial resolution BOLD fMRI by tailoring the k-space coverage to the spatial smoothing settings, permitting optimisation of the acquisition parameters. Lastly, an optimised pulse sequence design for fMRI data acquisition, namely Field Mapping Embedded EPI (FME-EPI), that improves B0 inhomogeneity-induced distortion is presented. This modified EPI pulse sequence is able to acquire both functional data and B0 inhomogeneity information concurrently, which can be used in EPI reconstruction to improve the spatial fidelity of the reconstructed images.
-
ItemTargeting seamless cartilage repair with a bioadhesive implant( 2022)Injury to articular cartilage in the knee can lead to post-traumatic osteoarthritis if untreated, causing debilitating problems later in life. Osteoarthritis impacts an individuals’ mobility, ability to work, and participation in daily activities. The broader impacts of this are significant, with osteoarthritis a leading cause of disability and an economic burden globally. Standard surgical treatments fail to ensure long lasting repair of damaged cartilage, often resulting in fibrotic tissue. The field of tissue engineering has seen a vast amount of research in cartilage regeneration, with few strategies reaching clinical trials. A common theme among failure of tissue engineered implants is their inability to integrate with the native tissue. Cartilage is a deceptively complex tissue despite its lack of innervation or blood supply. Its matrix is dense, heterogeneous and anti-adhesive, containing only a small number of cells and little ability for self-repair. This work seeks to understand if a cell-laden bioadhesive material can improve integration with cartilage, by bonding the regenerative implant to the native tissue. A novel bioadhesive comprised of photocrosslinkable gelatin methacryloyl and a biological enzyme, microbial transglutaminase, is reported. The material’s adhesion to cartilage ex vivo is assessed mechanically and chondrogenesis by human adipose derived stem cells (hADSCs) encapsulated within the material is evaluated. The enzyme significantly improved adhesion to cartilage ex vivo and did not impede the production of cartilage matrix by hADSCs cultured under chondrogenic stimulation conditions. In a preliminary study, the enzyme significantly improved integration with human cartilage explants over time under static culture conditions. The ability of the bioadhesive to support integration with cartilage ex vivo under cyclic compressive loading was then investigated, which is understood to be a first within the literature. No clear advantage of the bioadhesive was observed under loading, with good integration observed histologically in all conditions and a significant ten-fold increase in integration strength over the culture duration. This experimental model in combination with a biphasic finite element model allows future investigation of open questions in the field. Further work could see the bioadhesive material combined with other strategies to improve integration and long-term cartilage regeneration outcomes, providing an essential step towards clinical translation.
-
ItemSignal Analysis Techniques for Resting State Functional Near-Infrared Spectroscopy( 2022)Functional near-infrared spectroscopy (fNIRS) is widely used as a non-invasive neuroimaging modality. FNIRS signals measure the changes in oxygenated and deoxygenated haemoglobin concentrations in the cerebral cortex. Using fNIRS, functional brain activity can be estimated by measuring brain oxygenation and haemodynamics. Resting state functional connectivity (RSFC) evaluates the correlation of fNIRS signals between brain regions in the absence of tasks. FNIRS RSFC has been shown to be a powerful tool in analysing intrinsic and spontaneous brain activity in various research and clinical applications. The most commonly used method in evaluating fNIRS RSFC is the seed-based correlation method, which measures the linear association between a single brain region and other regions for which an fNIRS signal has been acquired. The strength of linear association is measured pairwise, considering each region separately, using Pearson's correlation coefficient, and tested for statistical significance. Inherently, implicit assumptions are imposed on the fNIRS signals when performing correlation-based connectivity analyses and subsequent tests of statistical significance. Pearson's correlation coefficients assume linearity, stationarity and Gaussianity of signals. The subsequent statistical tests impose an additional assumption that samples within the signal are independent of each other. Furthermore, any noise contained in the signals is assumed to be uncorrelated, so that sample correlation values are assumed to measure the linear association between cortical haemodynamic activity. Violation of any one of these assumptions may invalidate either the sample correlation estimate, or the statistical test from which conclusions are drawn regarding its statistical significance. The above-mentioned assumptions are violated by fNIRS signals. FNIRS are typically contaminated by noise and artefacts. Moreover, fNIRS signals are highly autocorrelated, due to fast sampling of the comparatively slow haemodynamic response of oxygenation changes in the brain. Furthermore, fNIRS signals are not stationary due to non-stationary components that may be distributed among fNIRS channels. Artefacts presented in the signals cause incorrect estimates of correlations in neural activity. In this thesis, a motion artefact removal algorithm based on robust estimation is proposed. Results show that the proposed algorithm successfully identifies and removes movement-related artefacts, and the performance of the algorithm is robust across different artefact conditions. A violated assumption of sample independence can alter RSFC estimates and invalidate consequent statistical hypothesis tests. To mitigate this, a statistical correction method is proposed to correct for correlation induced by non-white fNIRS frequency spectra and temporal filtering, thereby restoring validity to the statistical significance test results. The non-stationarity of fNIRS signals violates the stationarity assumption imposed by correlation. This thesis investigates the impact of multiplicative non-stationary signal components on fNIRS RSFC, and proposes a correction method to mitigate the impact of non-stationary multiplicative signal power. FNIRS RSFC can be evaluated with the proposed signal analysis methods, providing valid estimates of connectivity between spatially remote brain regions, and facilitating future applications of fNIRS.
-
ItemHigh-resolution macroscopic human connectomics( 2022)Recent advances in non-invasive neuroimaging have enabled the in-vivo acquisition of high-resolution brain imaging data and this, in turn, has unfolded a new chapter in systems neuroscience devoted to the study of human connectomics. Different imaging modalities allow us to map anatomical connections (white matter fiber tracts) as well as functional connections (synchronized activity) between pairs of brain regions. These tools provide an influential resource of noninvasively-captured human brain connectivity maps that portray the brain as a complex network of neuronal communications giving rise to unique individual variations associated with various behavioral characteristics and cognitive abilities. The principal goal of connectomics is to comprehensively map and study brain connectivity to better understand the structural and functional organization of the brain. In mesoscale, brain connectivity is normally mapped using a brain atlas that segments the brain into distinct regions serving as the nodes of the network. This atlas-based brain connectivity approach has been extensively studied in the past couple of decades and has proven advantageous in expanding our understanding of human brain connectivity organization. Nonetheless, simplification of the rich brain connectivity information acquired from MRI data into a coarse atlas-based connectome model may mask out interesting information about the human brain connectivity organization at a finer scale. This thesis explores a pioneering connectomic approach that aims to increase the spatial resolution of large-scale macroscopic brain-wide network models to better explore intricate maps of brain connectivity at higher spatial resolution. We term this novel approach "high-resolution connectomics". In this thesis, we explore the mathematical foundations of this novel method and provide computational tools enabling efficient mapping of high-resolution connectomes. We additionally present a wide range of empirical scenarios in which a higher-resolution connectome proves to be advantageous. Collectively, the work presented in this thesis forms the foundations of a new area of connectomics and showcases the various merits of this approach promoting its future use in human connectomic studies to advance our understanding of brain structure, function, and connectivity.