CMEM-based motor imagery induced cortical source localization for computationally efficient brain computer interface

Abstract

Localization of task-induced cortical sources plays a crucial role for an efficient design of brain computer interface (BCI) while reducing computational cost. Various entropy-based signal processing methods are being used to project actual cortical sources which could be utilized for low-cost and portable BCI. This paper empirically demonstrates a potential application of coherent-Maximum Entropy on the Mean (cMEM), a solution to the well-known electroencephalogram (EEG) inverse problem, in the context of classifying two motor imagery (MI) tasks, i.e., right hand and right foot, with a reduced number of channels involved. A paradigm of 10-fold cross validation, to train common spatial pattern (CSP) and regularized common spatial pattern (RCSP) for the estimation of spatially distinguishable features and to train support vector machine (SVM) classifier model, were used. The maximum classification accuracy of 93.21±3.56 has been achieved for a particular subject using only 17 channels and this is very close to 94.65±3.86 achieved with 118 channels. In addition, reduced number of channels requires lower computational time compared to 118 channels. Moreover, the effects of outliers on BCI performances can be reduced using only the optimal task-related channels. The results listed in this paper present the cMEM as a promising channel selection tool for efficient BCI systems.