A New Framework to Infer Intra- and Inter-Brain Sparse Connectivity Estimation for EEG Source Information Flow

摘要

This paper presents a new framework for sparse brain effective connectivity estimation of electroencephalographic (EEG) signals. The method is based on the sparsely connected source analysis using the time-varying multivariate auto-regressive (tv-MVAR) model, to find the intraand interbrain region connectivity. The proposed method shows the tv-MVAR model with sparse coefficient obtained from the least absolute shrinkage and selection operator (LASSO) penalty solver. To some extent, the sparse model has eliminated the false interpretation of connected brain region due to the presence of volume conduction effect, providing an in-depth understanding of brain network. The explicit contributions of this paper are: 1) newly introduced time-varying sparsed MVAR model (Adaptive autoregressive); 2) simplified sparsed framework to obtain the enhanced connectivity estimation; 3) sparsity constraint added by LASSO penalization method and its comparison with other penalization techniques; 4) comparison between the general non-sparsed tv-MVAR model and the sparsed tv-MVAR model; and 5) found to be a promising approach to locate the sparsely connected sources. The results are verified and found to be consistent with exhaustive real-time EEG time series obtained from the experiment during short guided meditation.