Low-Density EEG for Neural Activity Reconstruction Using Multivariate Empirical Mode Decomposition

Abstract

Several approaches can be used for estimating neural activity. The main differences between them are in the apriori information used and their sensibility to high noise levels. Empirical Mode Decomposition (EMD) has been recently applied to Electroencephalography EEG-based neural activity reconstruction to provide apriori time-frequency information to improve the neural activity estimation. EMD has the specific ability to identify independent oscillatory modes in non-stationary signals with multiple oscillatory components. The various attempts to use EMD in EEG analysis, however, did not provide yet the best reconstructions due to the intrinsic mode mixing problem of EMD. Some previous works have used a single-channel analysis and in other cases, multiple-channel have been used for other applications. In this paper, we present a study about multiple-channel analysis using Multivariate Empirical Mode Decomposition (MEMD) as a method to attenuate the mode mixing problem and to provide apriori useful time-frequency information to the reconstruction of neuronal activity using several low-density EEG electrode montages. The methods were evaluated over real and synthetic EEG data, in which the reconstructions were performed using multiple sparse priors (MSP) method with several electrode numbers of 32, 16, and 8, and the source reconstruction quality was measured using the Wasserstein Metric. Comparing the solutions when no pre-processing was made and when MEMD was applied, the source reconstructions were improved using MEMD as apriori information in the low-density montage of 8 and 16 electrodes. The mean source reconstruction error on a real EEG dataset was reduced a 59.42% and 66.04% for the 8 and 16 electrodes montages respectively, and on a simulated EEG with three active sources, the mean error was reduced an 87.31% and 31.45% for the same electrodes montages.