Classification Of Motor And Motor Imagery Based On High Frequency Characteristics Of Electrocorticogram

Motor and motor imagery can not only improve our motor skills,but also promote the rehabilitation of patients with hemiplegia.The research of brain computer interface（BCI）based on motor and motor imagery has important application value for paralyzed patients.At present,the research based on motor imagery mainly focuses on lowfrequency electroencephalogram（EEG）,but its spatial resolution is low,and the method of feature extraction is cumbersome.In addition,the classification effect among subjects is uneven.In recent years,electroencephalography（ECo G）has attracted extensive attention because of its high spatial resolution and high fidelity of high-frequency signals.Research shows that the change of high-frequency ECo G signal is specific to limb movement,so the characteristics of high-frequency ECo G can be used to classify motor or motor imagery.In this paper,76-200 Hz high-frequency energy is used to calculate the activation value of brain regions during motor or motor imagery,and explore the similarities and differences between them.At the same time,through the classification of motor or motor imagery tasks by ECo G high-frequency features,we explore whether the use of high-frequency features has advantages.The results obtained are as follows:（1）The results of brain activation analysis showed that both motor and motor imagery activated the primary motor cortex and auxiliary motor cortex.In addition,cognitive cortex was activated during motor imagery.At the same time,the overlap between the brain regions activated by hand and tongue is high in the case of motor imagery.The brain regions activated by each finger moving alone are similar,and the overlap of brain regions activated by middle finger,ring finger and thumb is high.（2）The results of time-frequency analysis of activated brain regions showed that event-related desynchronization in low frequency band and event-related synchronization in high frequency band of ECo G were found in activated brain regions.In addition,the time-varying characteristics of high-frequency energy in activated brain region have obvious specificity between hand and tongue movement.The specificity is poor when imagining hand and tongue movements.When five fingers move,high-frequency energy has obvious specificity between thumb and index finger.The specificity between middle finger,ring finger and little thumb is poor.（3）The classification results show that the classification results of common spatial pattern（CSP）features of phase-locked value network are affected by the overlap of activated brain regions of each task.The greater the overlap,the worse the classification effect.Using the CSP feature classification of specific frequency band can reduce the differences between subjects,and the average classification accuracy is greatly improved.In addition,the classification results of the two methods in the high frequency band are much better than those in the low frequency band.On the whole,the high-frequency CSP feature classification effect of ECo G is the best.The average accuracy of hand and tongue classification in the case of motor and motor imagery is 96.9% and 83.1% respectively.The average classification accuracy of five finger movements can reach 86.75%.Our results show that motor imagery can activate the primary motor cortex of the brain,and the process of motor imagery also needs the participation of cognitive nerves.In addition,the high-frequency CSP feature of ECo G can effectively carry out twoclassification and five-classification when the data set is very small.This simple and fast feature extraction method adopted in this paper has positive significance for the classification performance of BCI based on motor and motor imagery.