The analysis of decoding parameter selection of hand movements based on brain function network

In order to improve the noninvasive decoding precision of hand movement parameters from continuous electroencephalogram (EEG), the laws of the influence of decoding parameters such as brain rhythms and channel combination based on multiple linear regression models were investigated. Firstly, the wavelet coefficients of EEG characteristic frequencies corresponding to each channel were extracted. In addition, brain function network (BFN), which possessed characteristics of small-world network, was constructed based on the filter threshold on Network Cost and Spearman correlation coefficient matrixes of wavelet coefficients among the channels. Then, the influence of different task states on decoding parameters was studied and analyzed through parameters of BFNs (standard deviation of average degree, average path length, average clustering coefficient). At last, the decoding parameters of hand movement were singled out according to P value of Kruskal-Wallis. The results showed that EEG low and intermediate frequency bands and the 8 channels combination set have greater contribution to decode hand movement. The paper sheds light on new ideas for choosing decoding parameters of subsequent hand movements.