Enhanced Computation of the EEG-IC Polarities using a Genetic Algorithm
Jorge Munilla, Andrés Ortíz, Haedar Emad Sharef Alsafi, Juan Luis Luque · 2022 12th International Congress on Advanced Applied Informatics (IIAI-AAI) · 2022
Electroencephalogram (EEG) is nowadays an essential tool for clinical and cognitive brain research. ICA algorithms are applied to EEG recordings to find maximally temporally independent statistical source signals, which are further referred to as independent components (ICs) and which represent brain and non-brain processes. Clustering of IC topographies is a popular way to assess the results of EEG-based experiments and usually implies the computation of average images of such IC topographies. This computation, however, requires a previous process to determine the polarity of the different ICs because component scalp maps have not absolute polarity. This sign ambiguity problem is solved in the bibliography by using an image as a reference and setting the polarity of each IC according to the sign of the correlation of this IC with such reference. This method, however, has issues with low values of correlation. Thus, this paper introduces a different approach that addresses the polarity computation from a more general view, trying to find the polarity vector that minimizes the global error, and proposes the use of a genetic algorithm to find it. The results obtained with this approach outperform the two main methods currently used.