Research On Improving And Enhancing Decoding Of Motor Imagery Based Brain-computer Interface

Brain Computer Interface(BCI)enables communication between the brain and external devices,with broad potential applications in healthcare,education,and entertainment.Improving the performance of Motor Imagery(MI)-based BCI systems is currently a research focus,and the classification algorithm of EEG signals is an important means of improving performance.Traditional classification algorithms mostly use spatial features on global EEG channels to classify EEG signals,but in MI tasks,some EEG channels may be redundant,leading to information loss and an inability to reflect channel differences.To address this,a method of constructing local channels is proposed to solve the problem of channel information differences frequently encountered in BCI systems and improve classification performance.Functional electrical stimulation(FES)is another way to enhance motor imagery,and most current research focuses on qualitative analysis of FES in enhancing MI,with little research on how FES enhances MI and to what extent.To address this,a quantitative analysis method for the degree of MI enhancement under different levels of FES is proposed to analyze the physiological principles and the trend of enhancement with changes in electrical stimulation.The proposed feature extraction method constructs local channels based on Euclidean distance and jointly diagonalizes each local channel to obtain multiple local spatial features for feature classification.Compared with the top five algorithms in the international BCI competition dataset,the average Kappa values were higher by 0.010,0.013,0.059,0.060,and 0.270,respectively.Then a two-dimensional feature distribution scatter plot was drawn to verify the performance of the algorithm,and finally,a paired t-test was used to analyze significant differences between the proposed algorithm and the five competitive algorithms.The results showed that the proposed algorithm was significantly different from two of the algorithms,and the p-value was very close to 0.05 for the other three algorithms(when p is less than 0.05,it is considered a significant difference between two samples).The experimental results showed that the proposed algorithm had better classification performance compared to other algorithms.Secondly,a quantitative analysis method was designed to analyze the effect of MI enhancement under different levels of FES.Four upper limb MI databases(0Hz/5Hz/10Hz/15Hz)were established,and the classification accuracy was calculated,indicating that the overall classification accuracy of MI tasks with FES was higher than that of MI tasks alone.The average classification accuracy of the experiment with 15 Hz functional electrical stimulation reached about 75%,which was about 11% higher than the MI task without FES.The ETMs(event-related desynchronization/synchronization maps)under different FES levels were also studied.The comparison results showed that all four situations had event-related desynchronization(ERD)phenomena,but the impact of FES on the ERD phenomenon was more obvious,and became more pronounced with an increase in FES intensity.In summary,this paper not only proposes a local channel construction algorithm for EEG signal feature extraction to improve classification performance,providing an algorithmic foundation for the practical application of MI-based BCI systems,but also proposes a quantitative analysis method for FES in enhancing MI,providing an experimental paradigm for analyzing the role of FES in MI-based BCI.