| The rehabilitation of motor function is one of the important indexes ofrehabilitation of stroke patients with hemiplegia. The rehabilitation methods mainlyinclude drug therapy and passive rehabilitation, which have played a certain effect,but the recovery rate of stroke patients is still very low, which can’t reach the level ofsatisfaction to all. The study that brain-computer interface technology based on activeconsciousness is introduced in the rehabilitation training has become a hot researchtopic in the field of rehabilitation engineering. Feature extraction of motion-relatedEEG and availability analysis are the key problems of the closed-loop feedback.In order to extract effective motional information from motion-related EEG, thispaper firstly designed motion-related experiment. The experiment was performed byseven subjects that right hand moved to three directions (left, top, right) and EEG dataof reciprocating motion and hand motion trajectory were recorded. The motor taskincluded preparation, execution and back stage. Two kinds of data were synchronizedby pulse signal.Then the time-frequency characteristic of the motion-related EEG was analyzed.The wavelet analysis was applied to confirm the motion-related bands in frequencydomain that mainly concentrated in delta and theta bands. Therefore,(1-8Hz) powerspectral density, the wavelet coefficient as well as the joint features of the two ofpreparational and motion stage were respectively extracted from motion-related EEG,and support vector machine (SVM) is used to predict the direction. Based on featuresof single electrical, the recognition accuracies of19electrodes were calculatedrespectively. The recognition results were used to identify the availability of theextracted features. The results showed that the recognition accurac ies of joint featuresof power spectrum and wavelet coefficient identification were higher than singlefeature, which was up to62.7%, and the features of preparational stage were moreeffective than motion stage, resulting in better recognition results.In order to further improve the direction recognition accuracy and look for thebest combination of electrodes, the research that the features based onmulti-electrodes were used to recognize the direction of motion was studied. It wasfound that motor task caused active exchange of information between different brain regions, and that preparational stage was more strong than execution stage through themutual information complexity analysis. Brain causal network was obtained bygranger causality analysis. The in-degree of each electrode was counted by totalnumber of information flows that flowed into this eclectrode, which was used tomeasure the importance in the process of information transmission of whole brain.Using multi-electrodes that remove features of the electrode with the largest value ofin-degree one by one could result in higher recognizable accuracies which wereusually more than90%when the number of electrodes was between6and14. Theseresults illustrated that it was effective to improve the recognizable accuracies of motordirection with causal analysis algorithm.The results of research showed that there were enough features extracted from theEEG signal of upper limb motion which could effectively distinguish the direction ofthe motion and that optimal selection of more electrodes could significantly improverecognizable accuracies, and that optimal combinations of electrodes were differentfrom different subjects, which was the individual fitness. Especially, feature extractionof preparational stage and the accurate recognition of direction would lay thefoundation for the implementation of mechanical auxiliary closed-loop activerehabilitation system in the future. |
PDF Full Text Request