Title:Toward an improved BCI for damaged CNS-tissue patient using EEG-signal processing approach

Abstract:This article examined brain signals of people with disabilities using various signal processing methods to achieve the desired accuracy for utilizing brain-computer interfaces (BCI). EEG signals resulted from 5 mental tasks of word association (WORD), Mental subtraction (SUB), spatial navigation (NAV), right-hand motor imagery (HAND), and feet motor imagery (FEET) of 9 people with central nervous system (CNS) tissue damage were used as input. In processing this data, Butterworth band-pass filter (8-30 Hz) was used in the preprocessing, and CSP, TRCSP, FBCSP methods were used in feature extraction, and LDA, KNN, linear and nonlinear SVM were used in classification stages. The training and testing process was repeated up to 100 times, and the random subsampling method was used to select the training and test data. Mean accuracy in 100 replications was reported as final accuracy. The classification results of two classes of 5 mental tasks on the data of 9 people with central nerve damage showed that the combination of FBCSP with KNN classifier has the highest accuracy of 69+-1.1 which belongs to the two classes of word association and feet motor imagery. The study of different feature extraction methods and classification indicates that the proper feature selection method and especially classifier is crucial in accuracy rate. Furthermore, it is necessary to pay attention to successful signal processing methods in designing a brain-computer interface to use in people with central nervous system.