Enhanced Pattern Recognition-based And Simultaneous And Proportional Control For Myoelectric Human-machine Interface

Myoelectric control of multifunctional prostheses is a paradigm of human-machine interface(HMI)studies,which mainly has two orientations: pattern recognition-based myoelectric HMI and simultaneous and proportional control(SPC)for myoelectric HMI.For pattern recognition-based myoelectric HMI,though the classification error rate(CER)of 10 motions of an able-bodied subject under ideal experimental conditions can be less than 5%,the clinical use and commercial impact of pattern recognitionbased myoelectric prostheses are still limited.The main reasons for this situation are:1.the wide difference between the able-bodied subject and the amputee,i.e.the poor quality of electromyography(EMG)signals of amputees;2.the massive gap between ideal experimental conditions and daily usage environment of prostheses,i.e.electrode shift,fatigue and long-term use and so on(this dissertation focuses on electrode shift)causing variation of EMG signals in practical application,all of which would give rise to high CER.Regarding to SPC for myoelectric HMI,poor quality of EMG signals of naive amputees with no experience in SPC would result in poor online control performance.Last but not least,for partial-hand amputees,since their wrists were still functioning well,their wrist motion would impact the acquired EMG signals.Regarding to the four above mentioned problems,this dissertation considers enhancing the performance of myoelectric HMI from the “human”(i.e.signal source)and the “machine”(i.e.algorithms)respectively.Through transcranial direct current stimulation(tDCS),common spatial patterns(CSP)and stationary subspace analysis(SSA),based on dozens of experiments on amputees and able-bodied subjects,the following works are done:1.To improve the poor performance of pattern recognition-based myoelectric control for amputees,an enhanced pattern recognition-based myoelectric HMI through tDCS was proposed.This method utilized the characteristics that tDCS could improve the motor cortex excitability to enhance the quality of EMG signals and the performance of pattern recognition-based myoelectric control through tDCS of the primary motor cortex corresponding to the affected side.The proposed myoelectric HMI was tested on six amputees.The results demonstrated that tDCS significantly reduced the average CER by 10.1%.After tDCS,the CER could attain a usable level(<20%).It has potential in dramatically reducing the length of learning process of amputees for effectively using pattern recognition-based myoelectric prostheses.2.To improve the poor online SPC performance of naive amputees,an enhanced SPC for myoelectric HMI through tDCS was proposed.We tested the proposed experiments on six naive amputees.The results demonstrated that tDCS could significantly improve the online SPC performance to an acceptable level and would be an effective intervention to improve the online SPC performance in a short time.3.The electrode shift,which may occur during donning/doffing of the prosthetic socket,is one of the main reasons for the increase in CER.Multiclass CSP with two types of schemes,namely one versus one(CSP-OvO)and one versus rest(CSP-OvR),were used to realize the spatial filtering of high-density EMG signals and enhance the robustness against electrode shift for myoelectric control.We tested nine intactlimb subjects.With respect to three commonly used features(time-domain,TD,timedomain autoregressive,TDAR and variogram,Variog),the CSP features significantly enhanced the robustness against electrode shift for myoelectric control.This could serve as an enhanced pattern recognition-based myoelectric HMI through CSP.4.Regarding to partial-hand amputees with functional wrists,through combining the pattern recognition-based and SPC for myoelectric HMI,a switch regime,including a linear discriminant analysis(LDA)classifier and 14 state-space models,was proposed to continuously estimate the finger joint angles under different static wrist motions from EMG.We tested the proposed switch regime on six able-bodied subjects and two partial-hand amputees.The results showed that the proposed switch regime was effective for continuous estimation of the finger joint angles under different static wrist motions from EMG.Furthermore,a class-wise SSA(cwSSA)was proposed and significantly reduced the CER of the seven different wrist motions,and consequently enhanced the estimation performance of the proposed switch regime.In summary,this dissertation has,from the “human”(i.e.signal source)and the “machine”(i.e.algorithms)respectively,enhanced the performance of pattern recognition-based and SPC for myoelectric HMI,thus having potential to promote myoelectric HMI to clinical application.