Spurious correlations in simultaneous EEG-fMRI driven by in-scanner movement

Highlights

• Motion causes spurious effects using common artifact correction in EEG-fMRI analysis.

• Spurious motion effects resemble neurophysiological plausible effects.

• Minor task-related motion can cause spurious task-related EEG effects.

• Motion–BOLD and EEG–BOLD correlations are largely overlapping after convolution with the HRF.

Abstract

Simultaneous EEG-fMRI provides an increasingly attractive research tool to investigate cognitive processes with high temporal and spatial resolution. However, artifacts in EEG data introduced by the MR scanner still remain a major obstacle. This study, employing commonly used artifact correction steps, shows that head motion, one overlooked major source of artifacts in EEG-fMRI data, can cause plausible EEG effects and EEG–BOLD correlations. Specifically, low-frequency EEG (< 20 Hz) is strongly correlated with in-scanner movement. Accordingly, minor head motion (< 0.2 mm) induces spurious effects in a twofold manner: Small differences in task-correlated motion elicit spurious low-frequency effects, and, as motion concurrently influences fMRI data, EEG–BOLD correlations closely match motion-fMRI correlations. We demonstrate these effects in a memory encoding experiment showing that obtained theta power (~ 3–7 Hz) effects and channel-level theta–BOLD correlations reflect motion in the scanner. These findings highlight an important caveat that needs to be addressed by future EEG-fMRI studies.

Introduction

Simultaneous EEG and fMRI recordings provide an immensely useful neuroimaging technique as they offer the unique possibility to non-invasively record neural activity at the highest temporal and spatial resolution (Debener et al., 2006). Such rich multidimensional datasets allow for numerous ways of merging EEG and fMRI data (Huster et al., 2012) with the most popular approach being EEG-informed fMRI analysis. Here, EEG parameters of interest are used to create a model of the BOLD responses. BOLD signals can be correlated with ERP components (Debener et al., 2005), resting state alpha power (Goldman et al., 2002) or task-related oscillatory power changes (Hanslmayr et al., 2011). Resulting spatial maps provide regions in which BOLD signal changes correlate with EEG parameters indicating a common generator of BOLD and EEG signals.

However, the biggest restraining factor in simultaneous recordings is still the quality of the EEG data. Usually, two types of artifacts are considered: the gradient and ballisto-cardio-graphic (BCG) artifacts (Allen et al., 1998, Debener et al., 2008, Liu et al., 2012a, Mullinger et al., 2013, Niazy et al., 2005). Another major source of artifact, spontaneous movement, is rarely discussed. Motion is a general problem for simultaneous recordings, since movement of any conductive material (i.e. EEG electrodes and wires) in a static magnetic field (as in an MR scanner) causes electromagnetic induction and consequently an artifactual EEG signal. Therefore, even very minor head motion on a sub-millimeter level severely affects the EEG; for example, the tiny movements related to every heartbeat are visible in the EEG data as BCG (Debener et al., 2008, Mullinger et al., 2013). Most researchers employing EEG-fMRI accept that these artifacts remain to some extent in their data, even after careful preprocessing, implicitly assuming that those artifacts are mainly decreasing the signal to noise ratio but not introducing spurious effects. This logic fails when head motion is correlated with the task parameters of interest (i.e. paradigm, behavioral performance, BOLD signals). Since it is well known that fMRI–BOLD signals are correlating with motion (voluntarily or physiologically driven), BOLD-motion correlations might be a serious concern for EEG-fMRI correlations (Birn et al., 1999, Friston et al., 1996, Murphy et al., 2013, Power et al., 2014).

We here present data measuring EEG inside and outside an MR scanner. EEG and simultaneous EEG-fMRI data were recorded during a memory paradigm where we focused on theta oscillations and their relation to BOLD signals (Fig. 1A). The memory-relevant aspects of this dataset will be reported in detail elsewhere (Fellner et al., in preparation) and are only in so far relevant for this study as motion-induced spurious “memory-like” effects. Specifically, we demonstrate that: (i) in-scanner EEG data on electrode level is highly dominated by motion-related artifacts; (ii) task-related motion in-scanner can cause spurious task-related EEG power effects that are in stark contrast to artifact-free out-of-scanner data; (iii) in-scanner motion can drive spurious, but neurophysiologically plausible EEG–BOLD correlations.