Performance Enhancement And Optimization Methods For Multifunctional Prostheses Based On Multi-source Neural Decoding

The amputees are the largest minority in the world.Providing a dexterous and intelligent prosthesis for limb amputees is the major approach to restore the function of their lost limbs,which not only can improve the quality of their life,but also can reduce the burden of the country,the society and the family.Currently,the commercial myoelectric prostheses are based on a pair of agonist-antagonist muscles of residual arm to control one degree of freedom(DOF)of motion at a time.For individuals with transhumeral or shoulder-disarticulation amputations,whose disability is most sever,they must trigger a “mode switch” to sequentially select which DOF of motion is to be actuated.This type of operation is cumbersome and slow for amputees.Recently,Electromyography pattern recognition(EMG-PR)control method,an advanced and intelligent technique,have been investigated in many laboratories worldwide.In EMG-PR method,pattern recognition technology is used to decode the EMG signals and identify the user’s motion intent.And then the prosthesis will operate the identified motions.Since the EMG-PR methods can provide an intuitive and dexterous control of upper-limb prosthetic devices,it has become the research frontier and hotpot in the field of rehabilitation engineering.However,after decades of research efforts,the EMG-PR based multifunctional prostheses still has not been used in clinical application.The possible reasons are: the residual muscle is too less to provide enough EMG signals for the prosthesis control;and the classifier performance is usually susceptible to several interferences such as electrode size,electrode shift,force variation,and so on,which would greatly decay the stability performance of a trained EMG-PR classifier in identifying motion intentions.Therefore,this paper aims to work on the performance enhancement methods for multifunctional prostheses based on multi-source neural information.Firstly,this paper developed a motion-classification strategy based on EMG-EEG combination for high level amputations who had too less residual muscle to provide enough EMG signals.By using the proposed method,the motion classification performance of the classifier was increased more than 7%.Additionally,this method was optimized by channel selections in order to enhance the classification performance with a decreased number of electrodes.The optimal EMG and EEG acquisition positions were founded,which would provide important guidelines for the clinical application.Secondly,this paper proposed a feature filtering method to reduce the effect of interferences on the traditional EMG-PR method.By investigating the performance of filter with different type and different filter orders,we found the optimal filter parameters for able bodied subjects and amputees.Compared with those of the traditional EMG-PR method,the motion classification accuracy of the proposed method with feature filtering was increased 4.7% and 4.0% for able bodied subjects and amputees respectively.Thirdly,this paper investigated the effect of force variations on the prostheses control performance.And proposed two methods: a common spatial patterns(CSP)based motion classification method and a parallel classification strategy based on mean absolute value(MAV).Results showed that both of the two methods can effectively reduce the effect of the force variations on the motion classification.Lastly,this paper investigate the effect of unwanted motions(UMs)on the prostheses control performance.And proposed an extended sample strategy(ESS)to reduce the effect.Results showed that by using the ESS method,more than 80% of the UMs would be rejected by the classifier,greatly improving the stability of the trained classifier.To sum up,the current commercial prostheses cannot provide the intuitive and intelligent control for amputees.And the traditional EMG-PR based prostheses are limited to some subjective and objective issues such as less residual muscle and force variations,which cause the prostheses performance are not stability and reliability enough for clinical application.To solve these problems,this paper recruited some able bodied subjects and amputees and developed some performance enhancement methods for multifunctional prostheses based on multi-source neural information.The outcome of this paper might provide some useful solutions to improve the robustness of existing EMG-PR based motion classification method,promoting the multifunctional myoelectric prostheses advance toward the clinical application.