Hand Autonomous Motion Recognition Based On Cortico-muscular Coherence

Hand extension disorder is a common symptom in stroke patients.Combining the patient’s Electroencephalogram(EEG)and Electromography(EMG)with the assistant control of rehabilitation robot,clinical active rehabilitation medicine and robotics are integrated to replace the traditional rehabilitation therapists,which provides a new technical approach for the rehabilitation of stroke patients’ hand function.Rehabilitation training methods include active rehabilitation and passive rehabilitation.Compared with passive rehabilitation,active rehabilitation makes full use of patients’ autonomous consciousness and patients’ active participation.The main strategy of active rehabilitation is to use EMG to obtain the information of hand muscle strength.Stroke patients’ EMG is abnormally released due to the neurologic impairment,muscular atrophy,myasthenia and other symptoms.It is difficult to effectively detect the information associated with limb movements.EEG contains active consciousness information.However,the high nonstationarity of EEG makes it difficult to classify fine hand movements.Therefore,the combination of EMG and EEG is a technical approach to effectively detect different hand movements.The key is to extract and classify the features of EMG,EEG and Cortico-Muscular Coherence(CMC)signals,which represent the tension state of fingers.The aim is to determine the signals that can effectively reflect the autonomous action state of subjects.It provides an experimental support for exploring the recovery mechanism of hand motor function.Based on the data of EEG,EMG and CMC of healthy subjects under different hand tensions,this thesis explores an effective method to detect the information of the subjects’ autonomous movements.It provides a basis for the recognition of the autonomous movements of the hands based on the cortico-muscular coherence.The main research work includes:(1)In order to determine the signals that can effectively reflect the subjects’ autonomous action state,three experimental paradigms are designed to measure different tension states.EEG of 34 channels including FC5,FC4,FC1,C3,C4,C1 and EMG of Flexor Digitorum(FD)and Extensor Digitorum(ED)are collected simultaneously as the data basis of EEG and EMG analysis.(2)According to the energy distribution characteristics of EMG,the 10-100 Hz EMG band is intercepted and preprocessed as the effective band.Determining the effective EMG time interval under different tension states is the premise of effective feature extraction and classification.Moving average method is used to process the instantaneous energy of EMG,and threshold comparison is used to detect the effective EMG time interval.Absolute mean method is used to extract EMG features.The results show that there are differences under different tension states(resting state,tension 1,tension 2)of the hand.The Support Vector Machine(SVM)is used to classify the EMG data of BP4(ED channel)of 10 subjects under three tension states in two-level and multi-level classification.The accuracy of multi-level classification reached an average of 86.2%.(3)The key frequency bands are determined as the entry point,due to the high non-stationary and low signal-to-noise ratio of tension training EEG.Firstly,the internal and external noise of EEG is preprocessed,then the brain topographic sequence map is analyzed.The brain activation region is located,and 12 relevant channels such as FC5,C3 and CP3 are determined.Wavelet Packet Transform(WPT)combined with power spectrum method is used to determine the energy distribution of EEG in different frequency bands of the relevant channels,and the key frequency band of EEG is determined to be 20.02-20.51 Hz.(4)In order to identify EEG features in different tension training,the resting state and exercise state of EEG are identified as the entry point.Common Spatial Patterns(CSP)algorithm is used to extract the energy characteristics of three tension states.The results show that it is different among the energy characteristics of three tension states.Based on the multi-levelit CSP algorithm,SVM is used to classify the C3 channel β-band EEG under three tension states for 10 subjects.The accuracy of multi-level classification reached an average of 81.0%.(5)The effects of EEG,EMG and CMC-based autonomous consciousness signal detection on subjects’ autonomous action status are compared.Firstly,the EEG of each frequency band(δ,θ,α,β,γ)of EEG channel(C3,C5,CP3,P5,P3)with higher classification accuracy are coherently processed with EMG of two muscles(ED and FD)respectively.Then,the coherence coefficients of each frequency band are extracted and taken as feature parameters.The three tension states are classified and identified by SVM.The classification accuracy is an average of 84.4% by multi-level classification in C3 channel β band.The results of classification and identification of three kinds of tension states by three kinds of signals are compared and analyzed.The accuracy of multi-level classification and recognition of different tension states based on CMC signal is 3.4% higher than that only based on EEG.The accuracy of two-level classification and recognition of resting state-tension 1,resting state-tension 2 and tension 1-tension 2 increase by 1.6%,1.7% and 7.3% respectively compared to that only based on EEG.