| The research on exercise rehabilitation training and function evaluation of stroke patients is a hotspot in the artificial intelligence, neuroscience and rehabilitation engineering fields. The human machine interface based on electroencephalogram(EEG) or electromyogram(EMG) signal can assist patients control the external devices and improve the rehabilitation training by promoting neural plasticity. However, EEG and EMG signals are weak as well as complex and the patients may have problems of fatigue and insufficient muscle strength during movement, so the current rehabilitation system based on EEG or EMG control separately need to be further improved in accuracy and stability. In this paper, the fusion and synchronization analysis between the EEG and EMG signals were studied. The mutual complementary and coordination between the EEG and EMG signals was expounded to improve the accuracy and stability of motion recognition; Concurrently, the multi-level synchronization characteristics between the brain and muscles were analyzed to decode neurophysiological mechanism of hand movements; These researches provided theoretical basis for function evaluation and motor control in clinical rehabilitation.Firstly, the research and application backgrounds of EEG and EMG singals were summarized in the rehabilitation engineering field. The generation, collection and characteristics of the EEG and EMG signals were introduced and the general preprocessing methods for EEG and EMG were also described. Additonally, the detail preprocessing procedures were showed to remove artifacts in EEG and EMG signals.Secondly, in order to improve the accuracy of motion recognition, a novel strategy(PSO-SVM) for fusing the EEG and EMG signals was proposed to expound the mutual complementary and coordination features based on the support vector machine(SVM) and particle swarm optimization(PSO). In this method, the “fusion cofficient” was defined to fuse the EEG and EMG feartures, then the combination of the SVM and PSO methods were used to identify movement patterns and adaptively optimize “fusion cofficient”. Meanwhile, in order to improve the stability of motion recognition, the “fusion cofficient” was adjusted as the motion state varied, to reduce the impact of fatigue and insufficient strength on muscles.Thirdly, in order to explore the multi-level synchronization characteristics between the EEG and EMG signals, three synchronization methods, coherence, granger causality(GC) and transfer entropy(TE), were introdueced and analyzed to describe the linear, nolinear and complex characteristics in frequency domain. Furthermore, a novel method, named time-frequency transfer entropy(TF-TE), was proposed to analyze the nonlinear coupling strength and infromation flow between the EEG and EMG signals in time-frequency domain. Simulation anlaysis verified that the proposed methods was valid, and all these methods had different functions in describing the synchronization characteristics of different coupling models.Finally, experimental researches were carrided out to vertify the fusion strategy and synchronization analysis method between the EEG and EMG signals. The EEG over the scalp and EMG from the flexor carpi ulnaris(FCU) / extensor carpi radialis longus(ECRL) were recorded synchronously under wrist flexion or extension in 9 healthy subjects. The PSO-SVM model was applied to fuse EEG and EMG features and achieve effective classification of unilateral limb movement patterns. For synchronization analysis, the EEG and EMG data were collected simultaneously from 10 healthy subjects during maintaining four different static forces(10% MVC, 20% MVC, 40% MVC, 60% MVC). Then the methods of coherence, GC, TE and proposed TF-TE in this paper were applied to describe the linear and nonlinear corticomuscualr features in different time and frequency bands and different transfer directions, revealing the neural mechanisms that how the motor system controls the force output for hand movement. |
PDF Full Text Request