Trends in EEG-BCI for daily-life: Requirements for artifact removal

Abstract Since the discovery of the EEG principles by Berger in the 20’s, procedures for artifact removal have been essential in its pre-processing. In literature, diverse approaches based on signal processing, data mining, statistic models, and others compile information from multiple electrodes to build filters for artifact removal in the time, frequency or space domains. For almost one century, EEG acquisitions have required strict experimental conditions that included an isolated room, clinical acquisition systems, rigorous experimental protocols and very precise stimulation control. Under these steady experimental conditions, artifact removal techniques have not significantly evolved since then. However, in the last decade technological advances in brain-computer interfaces permit EEG acquisition by means of wireless, mobile, dry, wearable, and low-cost EEG headsets, with new potential daily-life applications, such as in entertainment or industry. New aspects not considered before, such as massive muscular and electrical artifacts, reduced number of electrodes, uncontrolled concomitant stimulus or the need for online processing are now essential. In this paper, we present a critical review of EEG artifact removal approaches, discuss their applicability to daily-life EEG-BCI applications, and give some directions and guidelines for upcoming research in this topic. Based on the results of the review, existing artifact removal techniques need further evolution to be applied in daily-life EEG-BCI. The use of multiple-step procedures is recommended, combining source decomposition with blind source separation and adaptive filtering, rather than using them separately. It is also recommendable to define and characterize most of artifacts evoked in daily-life EEG-BCI for a more effective removal.