Research On Hybrid Brain-Computer Interface Technology For Lower Extremity Exoskeleton Of Stroke Rehabilitation

Due to its high disability rate and high mortality rate,stroke has become the first factor in China's death and disability.Stroke rehabilitation is imminent.Under the demand of stroke rehabilitation,the combination of brain-computer interface and exoskeleton has become a hot topic in recent years.However,traditional brain-computer interfaces cannot cope with the requirements of exoskeleton for complex motion mode switching.The integration of a single brain-computer interface to form a hybrid brain-computer interface can well cope with this problem,and has been an important direction of brain-computer interface research in recent years.In the traditional brain-computer interface,the steady-state visual evoked potentials brain-computer interface can have a stable command output for the lower extremity exoskeleton.At the same time,the motor imagery brain-computer interface can provide a way for stroke patients to mental imagery.Therefore,based on steady-state visual evoked potentials and motor imagery,this paper studied hybrid brain-computer interface technology for lower extremity exoskeleton of stroke rehabilitation.The main research contents of this paper are as follows:(1)An identification method of steady-state visual evoked potentials electroencephalogram signals based on kernel canonical correlation analysis of incomplete Cholesky decomposition was studied.This method was mainly for the problem that the canonical correlation analysis was difficult to find the potentials correlation between the two sets of nonlinear correlation data.The kernel canonical correlation analysis was used to map both the stimulation frequency reference signal and the steady-state visual evoked potentials electroencephalogram signal into a kernel matrix of high dimensional space.Therefore,it found the hidden correlation between the two sets and selects the largest correlation coefficient among them as the recognition result.At the same time,it combined the incomplete Cholesky decomposition,eliminating unwanted interference information in the kernel matrix to reduce the size of the kernel matrix.It effectively accelerated the calculation of the kernel matrix and made the analysis of correlation more accurate.(2)A method for motor imagery electroencephalogram signal recognition and idle state detection in three types of common spatial pattern was studied.This method added the motor imagery idle state detection task to the two types of tasks of the left-hand motor imagery and the right-hand motor imagery.Therefore,this method introduced the feature extraction methods of three kinds of common spatial pattern:one versus one common spatial pattern and approximate joint diagonalization common spatial pattern,and classified and recognized three types of motor imagery tasks in the classification algorithm.(3)A hybrid brain-computer interface based on steady-state visual evoked potentials and motor imagery was studied,and its algorithm model has been implemented as an online software system.This method incorporated a finite state machine in the context of electroencephalogram signal recognition and idle state detection for steady-state visual evoked potentials and motor imagery,and encoded electroencephalogram signal recognition and idle state detection under different time windows.Thus,it was possible to switch between 7 types of lower extremity exoskeleton movement modes:stand,up the stairs,down the stairs,sit,walk,turn left,turn right.In this paper,the electroencephalogram signal recognition methods of steady-state visual evoked potentials and motor imagery were first tested on the offline datasets,and all had achieved good recognition results.Then,the online experiment of the hybrid brain-computer interface system showed that the hybrid brain-computer interface for lower extremity exoskeleton of stroke rehabilitation can provide a new type of rehabilitation pathway for stroke patients to some extent.