Research On EEG Feature Recognition Technique Based On Compound-limbs Motor Imagery

In recent years, the number of paralyzed patients caused by spinal cord injury hasincreased. How to recover and reconstruct the limb movement function effectively hasbecome a problem urgently to be solved in the field of rehabilitation engineering. Thebrain-computer interface based on motor imagery can develop an artificial neuralpathway close to the original damaged neural channel, which operation mode basedon endogenous induced response can output the user’s subjective movement-relatedconsciousness directly. With respect to motor imagery of simple limb movement,motor imagery of compound limb movement reflects mutual integration and synergiesbetween different functional areas of the brain, which is more close to the user’scustoms in daily life and, at the same time, also can satisfy the requirements ofmultiple instructions output of control information. So it has great significance for thelimb function rehabilitation after spinal cord injury.In order to explore the EEG feature induced by motor imagery of compound limbmovement combining hands with feet, seven kinds of motor imagery task mode weredesigned in this study, including left hand, right hand, both feet, both hands, left handcombined with right foot, right hand combined with left foot, and rest state. Toinvestigate the effect of train to task operation, two experimental stages have beendesigned:10subjects in the first stage untrained for motor imagery, and10subjects inthe second stage, including six from the first stage, trained for motor imagery a certaintime. Firstly, time-frequency spectrum, brain topographic map and power spectraldensity curve were used to analyze the time-frequency, energy feature and regiondistribution of event related desynchronization (ERD) induced by each motor imagerypattern qualitatively. The results of ERD value analysis on key channels showed thatthe ERD feature induced by compound motor imagery of hand movement combinedfoot movement was stronger than that induced by simple motor imagery of hand orfoot movement, and the ERD feature of every motor imagery mode was enhancedobviously after training. The results of Fisher separability analysis showed that thedifference between every two modes was amplified after training. Three kinds ofcommon spatial pattern algorithm including CSP，GECSP and sTRCSP were used forfeature extraction. The classification of seven kinds of motor imagery task modeinvolving hands and feet was done by support vector machine (SVM). The classification results showed that sTRCSP preformed better than the other twoalgorithms, and higher classification accuracy was obtained from the EEG signal aftertraining than that before training. The highest accuracy of binary classification couldreach99%, and the highest accuracy of seven-class recognition could reach84%.Multi-class compound motor imagery patterns combining hands with feet and itsfeature extraction methods from EEG signal are expected to provide techniqualsupport and helps to effectively expand the instructions of brain-computer interfacebased on motor imagery and further achieve the ideal goal of rehabilitation to "letthoughts into action" for the patients suffered from spinal cord injury.