Search

Neuromuscular diseases are a prevalent cause of prolonged and severe suffering for patients, and with the global population aging, it is increasingly becoming a pressing concern. To assess muscle activity in NMDs, clinicians and researchers typically use electromyography (EMG), which can be either non-invasive using surface EMG, or invasive through needle EMG. Surface EMG signals have a low spatial resolution, and while the needle EMG provides a higher resolution, it can be painful for the patients, with an additional risk of infection. The pain associated with the needle EMG can pose a risk for certain patient groups, such as children. For example, children with spinal muscular atrophy (type of NMD) require regular monitoring of treatment efficacy through needle EMG; however, due to the pain caused by the procedure, clinicians often rely on a clinical assessment rather than needle EMG. Magnetomyography (MMG), the magnetic counterpart of the EMG, measures muscle activity non-invasively using magnetic signals. With super-resolution capabilities, MMG has the potential to improve spatial resolution and, in the meantime, address the limitations of EMG. This article discusses the challenges in developing magnetic sensors for MMG, including sensor design and technology advancements that allow for more specific recordings, targeting of individual motor units, and reduction of magnetic noise. In addition, we cover the motor unit behavior and activation pattern, an overview of magnetic sensing technologies, and evaluations of wearable, non-invasive magnetic sensors for MMG.

IntroductionImproving the robustness of myoelectric control to work over many months without the need for recalibration could reduce prosthesis abandonment. Current approaches rely on post-hoc error detection to verify the certainty of a decoder's prediction using predefined threshold value. Since the decoder is fixed, the performance decline over time is inevitable. Other approaches such as supervised recalibration and unsupervised self-recalibration entail limitations in scaling up and computational resources. The objective of this paper is to study active learning as a scalable, human-in-the-loop framework, to improve the robustness of myoelectric control.MethodActive learning and linear discriminate analysis methods were used to create an iterative learning process, to modify decision boundaries based on changes in the data. We simulated a real-time scenario. We exploited least confidence, smallest margin and entropy reduction sampling strategies in single and batch-mode sample selection. Optimal batch-mode sampling was considered using ranked batch-mode active learning.ResultsWith only 3.2 min of data carefully selected by the active learner, the decoder outperforms random sampling by 4–5 and ~2% for able-bodied and people with limb difference, respectively. We observed active learning strategies to systematically and significantly enhance the decoders adaptation while optimizing the amount of training data on a class-specific basis. Smallest margin and least confidence uncertainty were shown to be the most supreme.DiscussionWe introduce for the first time active learning framework for long term adaptation in myoelectric control. This study simulates closed-loop environment in an offline manner and proposes a pipeline for future real-time deployment.

Muscles are the actuators of all human actions, from daily work and life to communication and expression of emotions. Myography records the signals from muscle activities as an interface between machine hardware and human wetware, granting direct and natural control of our electronic peripherals. Regardless of the significant progression as of late, the conventional myographic sensors are still incapable of achieving the desired high-resolution and non-invasive recording. This paper presents a critical review of state-of-the-art wearable sensing technologies that measure deeper muscle activity with high spatial resolution, so-called super-resolution. This paper classifies these myographic sensors according to the different signal types (i.e., biomechanical, biochemical, and bioelectrical) they record during measuring muscle activity. By describing the characteristics and current developments with advantages and limitations of each myographic sensor, their capabilities are investigated as a super-resolution myography technique, including: (i) non-invasive and high-density designs of the sensing units and their vulnerability to interferences, (ii) limit-of-detection to register the activity of deep muscles. Finally, this paper concludes with new opportunities in this fast-growing super-resolution myography field and proposes promising future research directions. These advances will enable next-generation muscle-machine interfaces to meet the practical design needs in real-life for healthcare technologies, assistive/rehabilitation robotics, and human augmentation with extended reality.

The study and measurement of the magnetic field from the skeletal muscle is called Magnetomyography (MMG). These magnetic fields are produced by the same ion currents which give rise to the electrical signals that are recorded with electromyography (EMG). For non-invasive measurements, the electric properties of subcutaneous tissue, i.e., most importantly, have a strong influence on the recorded signals. This paper presents a computational model to study the volume conduction effect with the finite-difference time-domain simulations using Sim4Life. The effects of 1 mm fat on the recorded electrical and magnetic signals from the skin surface have been evaluated in both EMG and MMG. The results indicate that due to 1 mm fat, the electrical signals decrease over 60% through traveling across layers between the muscle and skin surface, while these layers are transparent to the magnetic field. In a similar simulation procedure, when the new fibers are recruited, the interference among electrical signals makes the strength of recorded signals behave non-linearly proportional to the increasing number of active muscle fibers. Sim4Life simulations show that the recorded magnetic signals do not have the same trajectory as electrical signals. Hence, the changes in EMG signals caused by volume conduction effect can result in signal misinterpretations.

MagnetoMyoGraphy (MMG) is a method of studying muscle function via weak magnetic fields generated from human active organs and tissues. The correspondence between MMG and electromyography means directly derived from the Maxwell-Amp\`ere law. Here, upon briefly describing the principles of voltage distribution inside skeletal muscles due to the electrical stimulation, we provide a protocol to determine the effects of the magnetic field generated from a time-changing action potential propagating in a group of skeletal muscle cells. The position-dependent and the magnetic field behaviour on account of the different currents in muscle fibres are performed in temporal, spectral and spatial domains. The procedure covers identification of the fibre subpopulations inside the fascicles of a given nerve section, characterization of soleus skeletal muscle currents, check of axial intracellular currents, calculation of the generated magnetic field ultimately. We expect this protocol to take approximately 2-3 hours to complete for the whole finite-element analysis., Comment: To be submitted to Nature Protocols. 37 Pages and includes 10 figures

MagnetoMyoGraphy (MMG) with superconducting quantum interference devices (SQUIDs) enabled the measurement of very weak magnetic fields (femto to pico Tesla) generated from the human skeletal muscles during contraction. However, SQUIDs are bulky, costly and require working in a temperature-controlled environment, limiting wide-spread clinical use. We introduce a low-profile magnetoelectric (ME) sensor with analog frontend circuitry that has sensitivity to measure pico-Tesla MMG signals at room temperature. It comprises magnetostrictive and piezoelectric materials, FeCoSiB/AlN. Accurate device modelling and simulation are presented to predict device fabrication process comprehensively using the finite element method (FEM) in COMSOL Multiphysics®. The fabricated ME chip with its readout circuit was characterized under a dynamic geomagnetic field cancellation technique. The ME sensor experiment validate a very linear response with high sensitivities of up to 378 V/T driven at a resonance frequency of fres = 7.76 kHz. Measurements show the sensor limit of detections of down to 175 pT/Hz at resonance, which is in the range of MMG signals. Such a small-scale sensor has the potential to monitor chronic movement disorders and improve the end-user acceptance of human-machine interfaces.

Dictionary learning is a cornerstone of image classification. We set out to address a longstanding challenge in using dictionary learning for classification; that is to simultaneously maximise the discriminability and sparse-representability power of the learned dictionaries. Upon this premise, we designed class-specific dictionaries incorporating three factors: discriminability, sparsity and classification error. We integrated these metrics into a unified cost function and adopted a new feature space, i.e., histogram of oriented gradients (HOG), to generate the dictionary atoms. The rationale of using HOG features for designing the dictionaries is their strength in describing fine details of crowded images. The results of applying the proposed method in the classification of Chinese handwritten numbers demonstrated enhanced classification performance $(\sim98\%)$ compared to state-of-the-art deep learning techniques (i.e., SqueezeNet, GoogLeNet and MobileNetV2), but with a fraction of parameters. Furthermore, combination of the HOG features with dictionary learning enhances the accuracy by $11\%$ compared to the case where only pixel domain data are used. These results were supported when the proposed method was applied to both Arabic and English handwritten number databases.

Prosthetic hands can help people with limb difference to return to their life routines. Commercial prostheses, however have several limitations in providing an acceptable dexterity. We approach these limitations by augmenting the prosthetic hands with an off-the-shelf depth sensor to enable the prosthesis to see the object's depth, record a single view (2.5-D) snapshot, and estimate an appropriate grasp type; using a deep network architecture based on 3D point clouds called PointNet. The human can act as the supervisor throughout the procedure by accepting or refusing the suggested grasp type. We achieved the grasp classification accuracy of up to 88%. Contrary to the case of the RGB data, the depth data provides all the necessary object shape information, which is required for grasp recognition. The PointNet not only enables using 3-D data in practice, but it also prevents excessive computations. Augmentation of the prosthetic hands with such a semi-autonomous system can lead to better differentiation of grasp types, less burden on user, and better performance.

Machine learning-based myoelectric control is regarded as an intuitive paradigm, because of the mapping it creates between muscle co-activation patterns and prosthesis movements that aims to simulate the physiological pathways found in the human arm. Despite that, there has been evidence that closed-loop interaction with a classification-based interface results in user adaptation, which leads to performance improvement with experience. Recently, there has been a focus shift towards continuous prosthesis control, yet little is known about whether and how user adaptation affects myoelectric control performance in dexterous, intuitive tasks. We investigate the effect of short-term adaptation with independent finger position control by conducting real-time experiments with 10 able-bodied and two transradial amputee subjects. We demonstrate that despite using an intuitive decoder, experience leads to significant improvements in performance. We argue that this is due to the lack of an utterly natural control scheme, which is mainly caused by differences in the anatomy of human and artificial hands, movement intent decoding inaccuracies, and lack of proprioception. Finally, we extend previous work in classification-based and wrist continuous control by verifying that offline analyses cannot reliably predict real-time performance, thereby reiterating the importance of validating myoelectric control algorithms with real-time experiments.

Spintronic sensors, that are based on the tunnellingmagnetoresistive (TMR) effect, have been utilized in detecting low magnetic fields. However, still no computer-based model of these devices is available to integrated circuit designer to implement them in a hybrid spintronic-CMOS system. We developed a finite element method (FEM)-based model of a MgO-barrier TMR device in COMSOL Multiphysics®. The parameters of this model were extracted from the state-of-the-art fabrication and experimental data. Results were compared with respect to the model geometry and the used material. The proposed TMR sensor model offers a linear response with a high TMR ratio of 233% at 10 mV power supply. The model was exported to Cadence© Spectre to create a compact model using Verilog-A language. The developed sensor model was simulated with its analog front-end in same environment. This model provided a reliable benchmark for modelling of the future hybrid spintronic-CMOS developments.

Objective. The objective of this study was to compare the use of muscles appropriate for partial-hand prostheses with those typically used for complete hand devices and to determine whether differences in their underlying neural substrates translate to different levels of myoelectric control. Approach. We developed a novel abstract myoelectric decoder based on motor learning. Three muscle pairs, namely, an intrinsic and independent, an intrinsic and synergist and finally, an extrinsic and antagonist, were tested during abstract myoelectric control. Feedback conditions probed the roles of feed-forward and feedback mechanisms. Results. Both performance levels and rates of improvement were significantly higher for intrinsic hand muscles relative to muscles of the forearm. Intrinsic hand muscles showed considerable improvement generalising to decoder use without visual feedback. Results indicate that visual feedback from the decoder is used for transitioning between muscle activity levels, but not for maintaining state. Both individual and group performance were found to be strongly related to motor variability. Significance. Physiological differences inherent to the hand muscles can translate to improved prosthesis control. Our results support the use of motor learning based techniques for upper-limb myoelectric control and strongly argues for their utility in control of partial-hand prostheses. We provide evidence of myoelectric control skill acquisition and offer a formal definition for abstract decoding in the context of prosthetic control.

Objective. Recording of neural signals from intact peripheral nerves in patients with spinal cord injury or stroke survivors offers the possibility for the development of closed-loop sensorimotor prostheses. Nerve cuffs have been found to provide stable recordings from peripheral nerves for prolonged periods of time. However, questions remain over the design and positioning of nerve cuffs such that the separability of neural data recorded from the peripheral nerves is improved. Approach. Afferent electroneurographic (ENG) signals were recorded with nerve cuffs placed on the sciatic nerve of rats in response to various mechanical stimuli to the hindpaw. The mean absolute value of the signal was extracted and input to a classifier. The performance of the classifier was evaluated under two conditions: (1) when information from either a 3- or 16-channel cuff was used; (2) when information was available from a cuff placed either distally or proximally along the nerve. Main results. We show that both 3- and 16-channel cuffs were able to separate afferent ENG signals with an accuracy greater than chance. The highest classification scores were achieved when the classifier was fed with information obtained from a 16-channel cuff placed distally. While the 16-channel cuff always outperformed the 3-channel cuff, the difference in performance was increased when the 16-channel cuff was placed distally rather than proximally on the nerve. Significance. The results indicate that increasing the complexity of a nerve cuff may only be advantageous if the nerve cuff is to be implanted distally, where the nerve has begun to divide into individual fascicles.

Gaining ethical approval to conduct collaborative research in the NHS is transitioning. The process of gaining ethical approval from a Research Ethics Committee was viewed as cumbersome within the UK. This viewpoint fuelled continuous improvement of the approval system, most recently implemented by the Health Research Authority. This paper documents an 8-month process of gaining ethical approval, under Proportionate Review, to recruit NHS patients, by an academic biomedical engineering laboratory. Within the field of prosthetic healthcare technology, the multidisciplinary laboratory adopted co-design principles to establish an effective relationship with clinicians to collaboratively gain approval. The challenges and recommendations of acquiring such approval are presented in this paper, including time commitment, project management, strategic networks and stakeholder engagement. The authors believe that academic biomedical engineering research is ideally situated to collaborate with the healthcare sector. By forging a collaborative approach, biomedical engineering can lead in the translation of technology from early stage research to clinical adoption. This vision will be facilitated by gaining an understanding of the NHS ethical approval process. The paper provides insight to this regard, with the aim of informing academic, research management and healthcare domains of this transitioning field of work.

Humans excel in grasping and manipulating objects because of their life-long experience and knowledge about the 3D shape and weight distribution of objects. However, the lack of such intuition in robots makes robotic grasping an exceptionally challenging task. There are often several equally viable options of grasping an object. However, this ambiguity is not modeled in conventional systems that estimate a single, optimal grasp position. We propose to tackle this problem by simultaneously estimating multiple grasp poses from a single RGB image of the target object. Further, we reformulate the problem of robotic grasping by replacing conventional grasp rectangles with grasp belief maps, which hold more precise location information than a rectangle and account for the uncertainty inherent to the task. We augment a fully convolutional neural network with a multiple hypothesis prediction model that predicts a set of grasp hypotheses in under 60ms, which is critical for real-time robotic applications. The grasp detection accuracy reaches over 90% for unseen objects, outperforming the current state of the art on this task., Comment: ACCV 2018

ObjectiveHigh-Frequency alternating current (HFAC) nerve block has great potential for neuromodulation-based therapies. However nerve function recovery dynamics after a block is highly understudied. This study aims to characterise the recovery dynamics of neural function after an HFAC block.ApproachExperiments were carried out in-vivo to determine blocking efficacy as a function of blocking signal amplitude and frequency, and recovery times as well as recovery completeness was measured within a 0.7 s time scale from the end of block. The sciatic nerve was stimulated at 100 Hz during recovery to reduce error to within ±10 ms for measurements of recovery dynamics. The electromyogram (EMG) signals were measured from gastrocnemius medialis and tibialis anterior during trials as an indicator for nerve function.Main ResultsThe HFAC block was most reliable around 20 kHz, with block thresholds approximately 5 or 6 mA depending on the animal and muscle. Recovery times ranged from 20 to 430 milliseconds and final values spanned relative outputs from approximately 1 to 0.2. Higher blocking signal frequencies and amplitudes increased recovery time and decreased recovery completeness.SignificanceThese results confirm that recovery dynamics from block depend on blocking signal frequency and amplitude, which is of particular importance for neuromodulation therapies and for comparing results across studies using different blocking signal parameters.

Background Myoelectric pattern recognition systems can decode movement intention to drive upper-limb prostheses. Despite recent advances in academic research, the commercial adoption of such systems remains low. This limitation is mainly due to the lack of classification robustness and a simultaneous requirement for a large number of electromyogram (EMG) electrodes. We propose to address these two issues by using a multi-modal approach which combines surface electromyography (sEMG) with inertial measurements (IMs) and an appropriate training data collection paradigm. We demonstrate that this can significantly improve classification performance as compared to conventional techniques exclusively based on sEMG signals. Methods We collected and analyzed a large dataset comprising recordings with 20 able-bodied and two amputee participants executing 40 movements. Additionally, we conducted a novel real-time prosthetic hand control experiment with 11 able-bodied subjects and an amputee by using a state-of-the-art commercial prosthetic hand. A systematic performance comparison was carried out to investigate the potential benefit of incorporating IMs in prosthetic hand control. Results The inclusion of IM data improved performance significantly, by increasing classification accuracy (CA) in the offline analysis and improving completion rates (CRs) in the real-time experiment. Our findings were consistent across able-bodied and amputee subjects. Integrating the sEMG electrodes and IM sensors within a single sensor package enabled us to achieve high-level performance by using on average 4-6 sensors. Conclusions The results from our experiments suggest that IMs can form an excellent complimentary source signal for upper-limb myoelectric prostheses. We trust that multi-modal control solutions have the potential of improving the usability of upper-extremity prostheses in real-life applications. Electronic supplementary material The online version of this article (doi:10.1186/s12984-017-0284-4) contains supplementary material, which is available to authorized users.

The typical control of myoelectric interfaces, whether in laboratory settings or real-life prosthetic applications, largely relies on visual feedback because proprioceptive signals from the controlling muscles are either not available or very noisy. We conducted a set of experiments to test whether artificial proprioceptive feedback, delivered non-invasively to another limb, can improve control of a two-dimensional myoelectrically-controlled computer interface. In these experiments, participants’ were required to reach a target with a visual cursor that was controlled by electromyogram signals recorded from muscles of the left hand, while they were provided with an additional proprioceptive feedback on their right arm by moving it with a robotic manipulandum. Provision of additional artificial proprioceptive feedback improved the angular accuracy of their movements when compared to using visual feedback alone but did not increase the overall accuracy quantified with the average distance between the cursor and the target. The advantages conferred by proprioception were present only when the proprioceptive feedback had similar orientation to the visual feedback in the task space and not when it was mirrored, demonstrating the importance of congruency in feedback modalities for multi-sensory integration. Our results reveal the ability of the human motor system to learn new inter-limb sensory-motor associations; the motor system can utilize task-related sensory feedback, even when it is available on a limb distinct from the one being actuated. In addition, the proposed task structure provides a flexible test paradigm by which the effectiveness of various sensory feedback and multi-sensory integration for myoelectric prosthesis control can be evaluated.

The long-term stability and low-frequency composition of local field potentials (LFPs) offer important advantages for robust and efficient neuroprostheses. However, cortical LFPs recorded by multi-electrode arrays are often assumed to contain only redundant information arising from the activity of large neuronal populations. Here we show that multichannel LFPs in monkey motor cortex each contain a slightly different mixture of distinctive slow potentials that accompany neuronal firing. As a result, the firing rates of individual neurons can be estimated with surprising accuracy. We implemented this method in a real-time biofeedback brain–machine interface, and found that monkeys could learn to modulate the activity of arbitrary neurons using feedback derived solely from LFPs. These findings provide a principled method for monitoring individual neurons without long-term recording of action potentials., The use of local field potential (LFP) brain signals may allow development of more efficient and robust neural prosthetic devices. Here, Hall et al. develop a method for estimation and biofeedback control of single-neuron firing rates using signals extracted from multiple low-frequency LFPs.

Proceedings of the fifth International Brain-Computer Interface Meeting

In this paper, we elaborate on the distinction between classic approaches towards brain machine interface (BMI), that is Biomimetic and Biofeedback, and discuss their advantages and disadvantages. For biomimetic BMI, we briefly report results of a novel constrained Kalman filtering-optimization mechanism for prediction of myoelectric signals from neural spike recordings in a behaving macaque monkey. For the Biofeedback BMI, we review early works on operant conditioning and our recent results on emulating BMI by a myoelectric control interface.

Publication in the conference proceedings of EUSIPCO, Florence, Italy, 2006

Publication in the conference proceedings of EUSIPCO, Florence, Italy, 2006

The performance of intelligent electromyogram (EMG)-driven prostheses, functioning as artificial alternatives to missing limbs, is influenced by several dynamic factors including: electrode position shift, varying muscle contraction level, forearm orientation, and limb position. The impact of these factors on EMG pattern recognition has been previously studied in isolation, with the combined effect of these factors being understudied. However, it is likely that a combination of these factors influences the accuracy. We investigated the combined effect of two dynamic factors, namely, forearm orientation and muscle contraction levels, on the generalizability of the EMG pattern recognition. A number of recent time- and frequency-domain EMG features were utilized to study the EMG classification accuracy. Twelve intact-limbed and one bilateral transradial (below-elbow) amputee subject were recruited. They performed six classes of wrist and hand movements at three muscular contraction levels with three forearm orientations (nine conditions). Results indicate that a classifier trained by features that quantify the angle, rather than amplitude, of the muscle activation patterns perform better than other feature sets across different contraction levels and forearm orientations. In addition, a classifier trained with the EMG signals collected at multiple forearm orientations with medium muscular contractions can generalize well and achieve classification accuracies of up to 91%. Furthermore, inclusion of an accelerometer to monitor wrist movement further improved the EMG classification accuracy. The results indicate that the proposed methodology has the potential to improve robustness of myoelectric pattern recognition.

In this dissertation, advanced methods for electroencephalogram (EEG) signal analysis in the space-time-frequency (STF) domain with applications to eye-blink (EB) artifact removal and brain computer interfacing (BCI) are developed. The two methods for EB artifact removal from EEGs are presented which respectively include the estimated spatial signatures of the EB artifacts into the signal extraction and the robust beamforming frameworks. In the developed signal extraction algorithm, the EB artifacts are extracted as uncorrelated signals from EEGs. The algorithm utilizes the spatial signatures of the EB artifacts as priori knowledge in the signal extraction stage. The spatial distributions are identified using the STF model of EEGs. In the robust beamforming approach, first a novel space-time-frequency/time-segment (STF-TS) model for EEGs is introduced. The estimated spatial signatures of the EBs are then taken into account in order to restore the artifact contaminated EEG measurements. Both algorithms are evaluated by using the simulated and real EEGs and shown to produce comparable results to that of conventional approaches. Finally, an effective paradigm for BCI is introduced. In this approach prior physiological knowledge of spectrally band limited steady-state movement related potentials is exploited. The results consolidate the method.

This chapter summarises conversations between researchers working in healthcare and academia linked through membership of the Starworks Network, a UK National Institute for Health Research initiative to accelerate the translation of child prosthetics research into daily use. Specifically, it aims to unpack challenges identified by the network and critically analyse the current ‘state of the art’ in relevant upper limb myoelectric prostheses areas, informed by multiple perspectives. Each section outlines an area of emerging influence over the past decade which is likely to remain influential over the next. It begins with a brief introduction to the Starworks Network and concludes with recommendations from the authors Chapter Contents: • Starworks • 9.1 Co-design • 9.2 Additive manufacturing • 9.2.1 Open source • 9.2.2 Cost • 9.2.3 Bespoke fitting • 9.2.4 Individualisation • 9.2.5 Regulation • 9.3 Socket fit • 9.3.1 Digital manufacturing • 9.3.2 Adaptable sockets • 9.3.3 Johns Hopkins University • 9.3.4 Salford University • 9.4 Game-based training • 9.4.1 University of New Brunswick • 9.4.2 Medical University of Vienna • 9.4.3 Limbitless Solutions • 9.4.4 Newcastle University • 9.4.5 University of Groningen • 9.5 Recommendations • 9.5.1 Co-design • 9.5.2 Additive manufacturing • 9.5.3 Socket fit • 9.5.4 Game-based training • Acknowledgements • References.

The socket is an integral and important part of the prosthetic limb, providing link between body and technology. The quality of this connection must be considered as we strive towards embodiment of the prosthetic limb. The purpose of the socket is not just to provide a container for the residual limb but to provide a vessel where biomechanical forces can be transmitted from the body to the prosthetic componentry in the most energy -efficient manner while protecting the underlying tissues. The residual limb consists of bone surrounded by an envelope of soft tissues, including muscles which may no longer have insertion points. As the bone is able to move within the envelope of soft tissues, poor transmission of force and discomfort within the prosthesis can occur. Stabilising tissues to minimise bone movement within the socket is considered one of the primary goals in designing and constructing a well-fitting socket.

Upper-limb myoelectric prostheses are controlled by muscle activity information recorded on the skin surface using electromyography (EMG). Intuitive prosthetic control can be achieved by deploying statistical and machine learning (ML) tools to decipher the user’s movement intent from EMG signals. This thesis proposes various means of advancing the capabilities of non-invasive, ML-based control of myoelectric hand prostheses. Two main directions are explored, namely classification-based hand grip selection and proportional finger position control using regression methods. Several practical aspects are considered with the aim of maximising the clinical impact of the proposed methodologies, which are evaluated with offline analyses as well as real-time experiments involving both able-bodied and transradial amputee participants. It has been generally accepted that the EMG signal may not always be a reliable source of control information for prostheses, mainly due to its stochastic and non-stationary properties. One particular issue associated with the use of surface EMG signals for upper-extremity myoelectric control is the limb position effect, which is related to the lack of decoding generalisation under novel arm postures. To address this challenge, it is proposed to make concurrent use of EMG sensors and inertial measurement units (IMUs). It is demonstrated this can lead to a significant improvement in both classification accuracy (CA) and real-time prosthetic control performance. Additionally, the relationship between surface EMG and inertial measurements is investigated and it is found that these modalities are partially related due to reflecting different manifestations of the same underlying phenomenon, that is, the muscular activity. In the field of upper-limb myoelectric control, the linear discriminant analysis (LDA) classifier has arguably been the most popular choice for movement intent decoding. This is mainly attributable to its ease of implementation, low computational requirements, and acceptable decoding performance. Nevertheless, this particular method makes a strong fundamental assumption, that is, data observations from different classes share a common covariance structure. Although this assumption may often be violated in practice, it has been found that the performance of the method is comparable to that of more sophisticated algorithms. In this thesis, it is proposed to remove this assumption by making use of general class-conditional Gaussian models and appropriate regularisation to avoid overfitting issues. By performing an exhaustive analysis on benchmark datasets, it is demonstrated that the proposed approach based on regularised discriminant analysis (RDA) can offer an impressive increase in decoding accuracy. By combining the use of RDA classification with a novel confidence-based rejection policy that intends to minimise the rate of unintended hand motions, it is shown that it is feasible to attain robust myoelectric grip control of a prosthetic hand by making use of a single pair of surface EMG-IMU sensors. Most present-day commercial prosthetic hands offer the mechanical abilities to support individual digit control; however, classification-based methods can only produce pre-defined grip patterns, a feature which results in prosthesis under-actuation. Although classification-based grip control can provide a great advantage over conventional strategies, it is far from being intuitive and natural to the user. A potential way of approaching the level of dexterity enjoyed by the human hand is via continuous and individual control of multiple joints. To this end, an exhaustive analysis is performed on the feasibility of reconstructing multidimensional hand joint angles from surface EMG signals. A supervised method based on the eigenvalue formulation of multiple linear regression (MLR) is then proposed to simultaneously reduce the dimensionality of input and output variables and its performance is compared to that of typically used unsupervised methods, which may produce suboptimal results in this context. An experimental paradigm is finally designed to evaluate the efficacy of the proposed finger position control scheme during real-time prosthesis use. This thesis provides insight into the capacity of deploying a range of computational methods for non-invasive myoelectric control. It contributes towards developing intuitive interfaces for dexterous control of multi-articulated prosthetic hands by transradial amputees.

Neural prosthetic systems aim to assist patients suffering from sensory, motor and other disabilities by translating neural brain activity into control signals for assistive devices, such as computers and robotics prostheses, or by restoring muscle contraction through functional electrical stimulation (FES). In a neuro-motor prosthetic device, the prediction of intended muscle activity is required for effective FES. It has been already known that upper-limb electromyogram (EMG) signals in primates, can be accurately predicted during repetitive tasks, by decoding the spiking-activity (SA) of single cells and multi-unit activity (MUA) in the motor cortical areas. Recent work now suggests that EMG signals can also be decoded by local field potential (LFP) recordings from the same areas. In such decoding schemes, the number of input variables is usually very large and no systematic way of performing effective variable selection has yet been suggested. In this work, we demonstrated for the first time that muscle activity decoding from SA and LFP signals in the primary motor cortex (M1) and the ventral premotor cortex (PMv) areas is feasible during naturalistic free behaviour, and we compared the decoding performance of spike-, LFP- and hybrid decoders. We also tested the relative information in a number of LFP frequency bands, and found that mid-high and high-frequency bands (70 - 244 Hz) conveyed the most EMG-related information. Finally, we compared the decoding performance of a group of sparse regression algorithms, and we showed that a method based on the variational bayes (VB) outperformed the conventional Wiener cascade filter for LFP-decoders in the case of limited amount of training data. For longer training datasets, the results from all methods were comparable.