Hierarchical-Variational Mode Decomposition for Baseline Correction in Electroencephalogram Signals

Abstract

Electroencephalogram (EEG) signals being time-resolving signals, suffer very often from baseline drift caused by eye movements, breathing, variations in differential electrode impedances, movement of the subject, and so on. This leads to misinterpretation of the EEG data under test. Hence, the absence of techniques for effectively removing the baseline drift from the signal can degrade the overall performance of the EEG-based systems. To address this issue, this article deals with developing a novel scheme of hierarchically decomposing a signal using variational mode decomposition (VMD) in a tree-based model for a given level of the tree for accurate and effective analysis of the EEG signal and research in brain–computer interface (BCI). The proposed hierarchical extension to the conventional VMD, i.e., H-VMD, is evaluated for performing baseline drift removal from the EEG signals. The method is tested using both synthetically generated and real EEG datasets. With the availability of ground-truth information in synthetically generated data, metrics like percentage root-mean-squared difference (PRD) and correlation coefficient are used as evaluation metrics. It is seen that the proposed method performs better in estimating the underlying baseline signal and closely resembles the ground truth with higher values of correlation and the lowest value of PRD when compared to the closely related state-of-the-art methods.