Comparing Artifact Removal Techniques for Daily-Life Electroencephalography with Few Channels

This paper proposes a comparison between artifact removal techniques applied to real electroencephalographic data. The aim was to investigate the most suitable technique for artifact removal with a focus on wearability, portability, and low cost of the final system. A particular focus was thus put on the usage of few channels as a key feature to develop wearable and portable low-cost devices. Recent techniques relying on artifact subspace reconstruction or its Riemannian modification were considered along with more classical ones based on independent component analysis and principal component analysis. Different cut-off parameters were investigated in order to compare aggressive artifact removal to less aggressive one. The considered artifacts were divided into four categories: eye blinking, eye closing, eye moving, and muscle artifacts. Moreover, uncontaminated signal epochs were taken into account during the analysis for checking out if the artifact removal technique was affecting them too. The root means square error was exploited as the metric for assessing artifact removal. Results from three subjects suggest that artifacts subspace reconstruction is the most effective one, even when down to four channels are taken into account. Moreover, the results pave the way to the design of an hybrid technique to be applied when less than four channels are available for the analysis. Finally, optimization of the cut-off parameters should also be furtherly investigated.