Online Reduction of Artifacts in EEG of Simultaneous EEG-fMRI Using Reference Layer Adaptive Filtering (RLAF)

Brain Topogr DOI 10.1007/s10548-017-0606-7 ORIGINAL PAPER Online Reduction of Artifacts in EEG of Simultaneous EEGfMRI Using Reference Layer Adaptive Filtering (RLAF) David Steyrl1,4 · Gunther Krausz2 · Karl Koschutnig3,4 · Günter Edlinger2 · Gernot R. Müller‑Putz1,4 Received: 9 July 2017 / Accepted: 31 October 2017 © The Author(s) 2017. This article is an open access publication Abstract Simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) allow us to study the active human brain from two perspectives concurrently. Signal processing based artifact reduction techniques are mandatory for this, however, to obtain reasonable EEG quality in simultaneous EEG-fMRI. Current artifact reduction techniques like average artifact subtraction (AAS), typically become less effective when artifact reduction has to be performed on-the-fly. We thus present and evaluate a new technique to improve EEG quality online. This technique adds up with online AAS and combines a prototype EEG-cap for reference recordings of artifacts, with online adaptive filtering and is named reference layer adaptive filtering (RLAF). We found online AAS + RLAF to be highly effective in improving EEG quality. Online AAS + RLAF outperformed online AAS and did so in particular online in terms of the chosen performance metrics, these being specifically alpha rhythm amplitude ratio between closed and opened eyes (3–45% improvement), signal-to-noise-ratio of visual evoked potentials (VEP) (25–63% improvement), and VEPs variability (16–44% improvement). Further, we found that EEG quality after online AAS + RLAF is occasionally even comparable with the offline variant of AAS at a 3T MRI scanner. In conclusion RLAF is a very effective add-on tool to enable high quality EEG in simultaneous * Gernot R. Müller‑Putz 1 Laboratory of Brain‑Computer Interfaces, Institute of Neural Engineering, Graz University of Technology, 8010 Graz, Austria 2 GUGER TECHNOLOGIES OG, Graz, Austria 3 Department of Psychology, University of Graz, Graz, Austria 4 BioTechMed-Graz, Graz, Austria EEG-fMRI experiments, even when online artifact reduction is necessary. Keywords Simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) · Artifact reduction · Reference layer adaptive filtering (RLAF) · Online processing Introduction Non-invasive neuroimaging techniques offer the unique opportunity to investigate the active human brain without surgery. The two most popular non-invasive neuroimaging techniques are electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) (Michel and Murray 2012; Norris 2006). EEG measures electrical brain activity, whereas fMRI measures blood oxygenation level changes in the brain (Niedermeyer and Lopes da Silva 2005; Ogawa et al. 1990). These two techniques have partly complementary properties. For example, the time resolution of EEG is in the millisecond range, whereas it is in the range of seconds for fMRI. A second example is the spatial resolution of the techniques, which is commonly in the range of millimeters for fMRI and in the range of centimeters for EEG (Laufs 2012; He et al. 2011). The combing of EEG and fMRI was proposed to benefit from the best of both worlds. The combined simultaneous application of these two techniques allows comprehensive studies of the same brain activity from the electrophysiological and from the metabolic/vascular point of view. Examples of such studies include the combined or joint analysis of EEG and fMRI data such as e.g. in EEG-informed fMRI, the localization of transient brain activity, and also the analysis of the interaction of electrophysiology and metabolism (Huster et al. 13 Vol.:(0123456789) Brain Topogr 2012; Uludag and Roebroeck 2014; Debener et al. 2006; Ritter and Villringer 2006). This combination is often referred to as simultaneous EEG-fMRI. Unfortunately, these two techniques influence each other and deteriorate the data quality of the respective other. The additional EEG equipment inside the MRI scanner interferes with the static magnetic field and with the radio frequency field of the scanner. This interference generates field inhomogeneities and signal losses, which in turn degrade the fMRI data quality. Studies demonstrate that the data quality loss in fMRI varies between negligible and severe, but is never prohibitive (Bonmassar et al. 2010; Luo and Glover 2012; Jorge et al. 2015a). The effect of fMRI data acquisition on the EEG data quality is however critical (Mulert and Lemieux 2010; Mullinger and Bowtell 2011a). Over the past years, a variety of fMRI related artifacts in EEG of simultaneous EEG-fMRI have been described. Below, we give an overview, sorted by the usual magnitudes of the artifacts. The most prominent artifact is the so-called gradient artifact (GA), sometimes also referred to as the scanner artifact (Allen et al. 2000). It has amplitudes up to 1000 times higher than the EEG (Allen et al. 2000; Mullinger et al. 2011b). The switching of the scanner gradient during fMRI data acquisition causes this artifact by electromagnetic induction in the leads of the EEG electrodes. It repeats whenever a new volume acquisition starts. Although techniques to reduce this artifact are known, it is not possible yet to avoid it completely (Mullinger et al. 2011b; Jorge et al. 2015a; Assecondi et al. 2016). Various signal processing based methods have thus been developed to reduce the impact of this artifact. Average artifact subtraction (AAS) is one of them and probably the most widely used one (Allen et al. 2000). AAS exploits the repetitive and deterministic nature of the GA. A separate artifact template is compiled for each single artifact epoch of each EEG channel by averaging over adjacent epochs. This template is subsequently subtracted from the EEG. By averaging over adjacent artifact epochs, AAS can cope with slow changes of the GA, but not with brisk changes, due to e.g. motion of the study participant. Hence, although AAS reduces the GA largely, residuals of the GA are still present and they can be in the same order of magnitude as the EEG. Reducing the GA unveils a second artifact, the pulse artifact (PA), which is repetitive with the cardiac-pulse cycle. PA amplitudes have the same order of magnitude as the EEG amplitudes and they increas with the strength of the static magnetic field (Allen et al. 1998; Debener et al. 2007, 2008). The PA itself is mainly caused by motion of EEG electrodes, due to cardiac-pulse driven head nodding and due to expansion of blood vessels below the respective EEG electrode (Bonmassar et al. 2002). A second contributor to this artifact is voltage induction in EEG electrodes due to the acceleration of blood below the electrode. 13 Blood is electrically conductive and therefore surrounded by an electromagnetic field, when accelerated in a static magnetic field. The proportion of this second contributor is relativel (...truncated)