Development Of Upper-limb Rehabilitation System For Stroke Patients Based On Motor Imagery

The goal of this thesis is to develop a real-time online Brain Computer Interface(BCI)system based on motor imagery,which targets at rehabilitation exercise of stroke patients.This system mainly consists of three parts,motor imagery detection algorithm,rehabilitative exoskeleton and rehabilitation software platform,which collect the online EEG signal to detect the motion intention of patient then to control the exoskeleton helping move the impaired arms of patients.And in this way the active rehabilitation treatment can be achieved.The neurophysiology basis of this system is neuroplasticity theory,which postulates that motor imagery with the motion feedback from the exoskeleton can facilitate the impaired motion conductive path restored or reconstructed by making parts of dormant synapses awake and compensate to achieve the recovery.In this thesis the generating mechanism and characteristics of EEG signal are discussed to help design the processing method of EEG signal,including handling the artifacts such as EOG and EMG and adoption of spatial filter to increase the spatial resolution of EEG signal.Then the experiment procedure of motor imagery is introduced,including online and offline part,and the EEG signal datasets of respective motion of left and right hands are extracted and presented.Two feature extraction techniques are tested in this thesis: frequency domain features and spatial features.The power spectrum density(PSD)based on the AR model are tested in the frequency feature extraction,based on which some new features according to the average and varied characteristics of power spectrum are raised.Then the genetic algorithm are applied to select the best features which enhances the robustness of the algorithm.While for spatial features the common spatial pattern(CSP)are tested.Ultimately the frequency domain features are chosen to be the final features based on the classification results.For the classification design,linear discriminant analysis(LDA)and support vector machine(SVM)are tested.All algorithms are tested with different parameters on the training set using cross-validation and then LDA is verified to be the best algorithm on the final test set.In the exoskeleton design,three parts are talked about in this thesis,including structure design,hardware design and software design.The exoskeleton receives the control signal from the motor imagery measurement module through TCP to control the rehabilitation exercise of impaired arms of patients.The rehabilitation software platform is used to manage the whole process during a recovery training trial,including the user interaction,database management,rehabilitation process management,exoskeleton control and algorithm implementation.The research in this thesis is conducive to the design of rehabilitation system based on motor imagery,especially for the exploration of increasing the adaptation and robustness of motor imagery BCI system.The viability of motor imagery BCI system in the rehabilitation exercise of cerebral apoplexy patients is proved in this article,which places some foundation of future clinical application.