Coefficient-of-variation-based Channel Selection with a New Testing Framework for MI-based BCI
Xiao, Ruocheng,Huang, Yitao,Xu, Ren,Wang, Bei,Wang, Xingyu,Jin, Jing
DOI: https://doi.org/10.1007/s11571-021-09752-4
2021-01-01
Cognitive Neurodynamics
Abstract:In the motor-imagery (MI) based brain computer interface (BCI), multi-channel electroencephalogram (EEG) is often used to ensure the complete capture of physiological phenomena. With the redundant information and noise, EEG signals cannot be easily converted into separable features through feature extraction algorithms. Channel selection algorithms are proposed to address the issue, in which the filtering technique is widely used with the advantages of low computational cost and strong practicability. In this study, we proposed several improved methods for filtering channel selection algorithm. Specifically, based on the coefficient of variation and inter-class distance, a novel channel classification method was designed, which divided channels into different categories based on their contribution to feature extraction process. Then a filtering channel selection algorithm was proposed according to the previous classification method. Moreover, a new testing framework for filtering channel selection algorithms was proposed, which can better reflect the generalization ability of the algorithm. Experimental results indicated that the proposed channel classification method is effective, and the proposed testing framework is better than the original one. Meanwhile, the proposed channel selection algorithm achieved the accuracy of 87.7% and 81.7% in two BCI competition datasets, respectively, which was superior to competing algorithms.