Research On Semg Feature Extraction Algorithm And System Design For Virtual Rehabilitation

The research on rehabilitation and evaluation of patients after stroke with upper extremity motor dysfunction has become a hot area in modern rehabilitation engineering. While newer than other traditional rehabilitation therapies, virtual rehabilitation is growing with the benefits of friendly human-computer interaction and high flexibility. Such virtual rehabilitation has improved the conventional one-to-one therapy mode and directed the rehabilitation therapy onto a formulized, digital and handy way. However, there exist defects of operation security, training motivation, individual adaptability and validity of rehabilitation evaluation in classical virtual training systems for rehabilitation, which become limiting factors in their clinical application. In view of the above mentioned problems, surface electromyography(s EMG)-based analysis and biofeedback control were introduced into the virtual rehabilitation system in this paper. The research was aimed at presenting effective methods of s EMG parameters identification and dynamic features extraction which can reflect the variation of s EMG related to motion pattern. Meanwhile, also explored were the many characteristics for estimating the neuromuscular functional status from the perspective of EMG in order to quantitatively evaluate the recovery condition. Moreover, the s EMG-based control, evaluation and virtual reality were combined and applied to establish an upper limb rehabilitation system with the ability of intelligence, humanization and self-adaption. The main works are conducted as follows:Firstly, to address the non-stationary and nonlinear characteristics of s EMG signals, a novel single-channel feature extracting approach named local band spectral entropy based on wavelet packet was proposed and applied. Then nonnegative tensor factorization was used to extract multi-channel features of s EMG which retained the information of timefrequency-space domain. Consulting with the present s EMG analysis method, this work compared the performance and application of different algorthms and determined that both of the above two methods can effectively identify the movement intention with high classification accuracy.Secondly, focusing on solving quantitative evaluation on motor function states, three s EMG approaches based on clinical assessment techniques were purposed from different aspects. The reflex electromyographic threshold method was put forward to quantitatively evaluate the upper-limb spasticity of stroke patients. Nonnegative matrix factorization technique was used to extract the muscle activation patterns and muscle synergies during motions. An effective intermuscular coherence analysis method was studied to explore the neuromuscular oscillation and the pathomechanism of motor dysfunction. Furthermore, these methods were brought into the rehabilitation system and designed individualized rehabilitation training strategy.Thirdly, a virtual rehabilitation training and evaluating system was proposed one the basis of s EMG-based analysis and biofeedback technology. Aiming to help the patients conduct training actively, s EMG analysis and pattern recognition were conducted on the system and used as input commands to control the virtual environment. And evaluation based on s EMG was also introduced for quantitatively evaluating and timely tracking motor function state and recovery condition. Furthermore, the parameter optimization and individualized rehabilitation system based on s EMG-based control, s EMG-based evaluation and virtual reality techniques with different virtual environments for different rehabilitation stages was developed using C# language.Finally, experimental study about our rehabilitation training and evaluating system was conducted. Stroke patients and healthy subjects were enrolled in our experiments. Three scientific issues were researched here, which include analysis of the recognition accuracy of s EMG-based motor control, the reliability of s EMG-based motor function evaluation, and the effect of gradually rehabilitation training with this system, to verify the feasibility and effectiveness of the rehabilitation system designed in this paper. The research will provide new methods to evaluate motor funtion during neural rehabilitation such as stroke and lay the foundation for human-centered rehabilitation training strategy.