Towards Robust And Reliable Pattern Recognition Based Methods For The Control Of Upper Limb Multifunctional Prostheses

Upper limb amputation constitutes one of the main factors affecting the life style of amputees during activities of daily living(ADL).It basically places psychological,physical,and emotional burdens on amputees,thus limiting their productivity during ADL especially when they have to perform tasks that require their lost limb functions.More so,recent surveys have reported that the number of individuals with upper limb loss will continue to be on the increase over time,and this may negatively affect the families of the amputees,their immediate environments,their countries,and the world at large.As part of efforts to re-integrate upper limb amputees back into the society,prosthetic devices have been developed to help restore their lost arm functions.In this regard,upper limb prostheses with different types of control mechanisms including the Body-powered,Amplitude-based,and Electromyogram pattern recognition(EMG-PR)based methods have been proposed amongst others.Recently,EMG-PR based control method has been repeatedly reported as a potential strategy since it can provide intuitively dexterous control and support multiple degrees of freedom movements.After about a decade of concerted efforts from the academia and industry towards advancing EMG-PR control method,the currently available EMG-PR based multifunctional prostheses are not yet clinically viable and thus their acceptance rate is limited amongst upper limb amputees.The following are some possible factors affecting the clinical robustness of EMG-PR based prosthetic control: a)the mobility of subject that would cause changes in the EMG signal patterns when eliciting identical limb motions in mobile scenarios;b)the existing feature extraction methods which provides limited neural information for limb movement intent identification;c)inability of some amputees to generate sufficient EMG signals from which their limb movement intentions could be decoded.Therefore,this dissertation investigated these issues and proposed a number of potential solutions to resolve them,as reported below.Firstly,the effect of mobility on the performance of EMG-PR motion classifier based on electromyogram and accelerometer signals acquired from six upper-limb amputees across four scenarios(one static and three mobile scenarios)was studied.Thereafter,three different solutions were proposed to mitigate the effect of mobility on EMG-PR based control strategy.By applying the proposed methods,the degradation in classification performance was significantly reduced from 8.98% to 1.86%(Dual-stage sequential method),3.17%(Hybrid strategy),and 4.64%(Multi-scenario strategy).Hence,this study may provide potential insight on improving the clinical robustness of the currently available multifunctional prosthesesSecondly,this dissertation proposed three new time-domain feature extraction methods with an attempt to improve the accuracy and robustness of EMG-PR prosthetic control.Experimental results obtained with EMG dataset from eight upper limb amputees showed that the proposed feature extraction method could achieved an average decoding accuracy of 92.00%±3.11% which was 6.49% higher than that of the commonly used time-domain features(p<0.05).With three additional evaluation metrics the proposed feature extraction method also performed better and this suggest that the new EMG features may facilitate the clinical realization of multifunctional.Thirdly,towards developing a multifunctional prostheses for amputees with neuromuscular disorder/high level amputation,32 EEG feature extraction methods(including 12 spectral-domain descriptors(SDDs)and 20 time-domain descriptors(TDDs))were used to decode multiple classes of motor imagery patterns associated with different upper limb movements based on 64-channel EEG recordings.With a linear combination of features from the individual domain,an optimal set of combined TDDs recorded an average accuracy of 90.68% while that of the SDDs achieved an accuracy of 99.55% which were significantly higher than those of the individual TDD and SDD at p<0.05.The findings suggest that optimal feature set combination would yield a relatively high accuracy that may facilitate the practical development of a robust multifunctional neuroprosthesis.Finally,this dissertation identified and studied some critical issues challenging the clinical and commercial success of EMG-PR based multifunctional prostheses as it relate to the accuracy,stability,and reliability of the assistive device.Additionally,this dissertation provided potential solutions to the problems as well as insight on how to improve the overall performance of the currently available prostheses,and it also provided possible future research directions.