| Upper limb amputation is the complete or partial loss of the arm due to trauma,cancer,infection,blood vessel disease,and birth deformities or diseases.Arm amputation leads to a life-changing condition that hinders amputees from performing essential tasks(such as eating,drinking,dressing,and object manipulation)during daily life activities.Therefore,developing prosthetic devices that can help amputees regain their lost limb functions,thus re-integrating them into society,has been a hot research topic in the academia and industry over the years.In the recent years,remarkable advancements have been made in upper limb prostheses technology,particularly with respect to its design,socket fabrication,fitting techniques,suspension systems,and control schemes.Meanwhile,the control scheme largely dictates the extent to which the device can naturally restore lost limb functions,and pattern recognition(PR)based strategy has been identified as a potential control method because it can support intuitive actuation and provide multiple degrees of freedom functions,among other benefits.Over the years,concerted efforts have been made towards advancing PR-control schemes.However,prostheses with functionalities that could meet the overwhelming majority of upper limb amputees’ expectations(especially in terms of support for finemotor function and robustness against unavoidable multiple dynamic factors encountered during daily usage)constitute a major issue to date.This dissertation investigated core issues affecting the existing PR-controlled prostheses in practical settings and proposed corresponding solutions that includes;(a)resolving the co-existing effect of co-founding factors on the control performance of myoelectric prostheses,(b)developing a systematic framework for appropriate selection of feature set and windowing parameters for efficient motion intent characterization in the context of PR-based control scheme,(c)proposing a robust highdensity surface electromyogram(HD-s EMG)signal denoising method for improved classification performance and reconstruction of distinct and repeatable muscle activation patterns and,(d)proposing a linearly extendable EEG artifacts removal algorithm for enhancing MI tasks decoding in transhumeral subjects and feasibility of practical application of the proposed method.The first study systematically investigated the co-existing impact of subject mobility(Mo S)and muscle contraction force variation(MCFV)on the performance of the PR-based movement intent classifier,using EMG recordings acquired from eight participants who performed multiple classes of targeted limb movements across static and non-static scenarios with three distinct muscle contraction force levels.Afterwards,a robust feature extraction method that is invariant to the combined effect of Mo S-MCFV,namely,invariant time-domain descriptor(inv TDD),was proposed to optimally characterize the multi-class EMG signal patterns in the presence of both factors.The study’s outcome consistently showed that the proposed inv TDD method could significantly mitigate the co-existing impact of Mo S-MCFV on PR-based movement-intent classifier with error reduction in the range of7.50%～17.97%(p<0.05),compared to the commonly applied methods.The second study examined the interaction effects of feature extraction methods and windowing parameters on the EMG-PR system’s performance towards constructing optimal parameters for accurately movement intent decoding in the context of prosthetic control.The study’s findings show that multiple feature framework consistently achieved minimum decoding error below 10% across optimal windowing parameters of 250 ms/100 ms,compared to a single feature framework.Also,multiple feature framework showed high robustness to additive noise with obvious trade-offs between accuracy and computation time.In the third study,due to poor decoding performance of motion intents in above-elbow amputees as a result of the noises contained in the myoelectric signals,a signal denoising method driven by multiscale local polynomial transform(MLPT),with the capability to improve HD-s EMG signal quality was proposed.The experimental results demonstrated that the proposed method outperformed the considered existing methods for various motion task decoding with over 13.0% increment in accuracy across subjects.The fourth study proposed a novel electroencephalogram(EEG)preprocessing method that combines Generalized Eigenvalue Decomposition(driven by low-rank approximation),and a Multichannel Wiener Filter(that employs a learning technique)to simultaneously eliminate multiple artifacts from motor imagery EEG(MI-EEG)recordings from four transhumeral subjects.The outcome revealed that the proposed method yields significant improvements in MI task decoding accuracies,in the range of 13.23%-41.21% compared to four existing popular artifact removal algorithms.Further investigation revealed that the method achieved accuracies in the range of 90.44%-99.67% using "single trial" EEG recordings,which could eliminate the need to record and process large ensembles of EEG trials as commonly required in existing approaches.Additionally,using a variant of the sequential forward floating selection algorithm,a subset of 9 channels was used to obtain a decoding accuracy of 93.69%.In summary,this dissertation’s findings offer compelling insight into developing a reliable control scheme for multifunctional prostheses that would be clinically viable.It can also provide proper insight for appropriate parameter selection in the context of robust PR-based control strategy for intelligent rehabilitation devices.In addition,the method proposed in this dissertation may potentially spur the development of effective real-time control strategies for multi-degree of freedom miniaturized rehabilitation robotic interfaces. |
