Elsevier

NeuroImage

Volume 191, 1 May 2019, Pages 21-35
NeuroImage

Investigating the variability of cardiac pulse artifacts across heartbeats in simultaneous EEG-fMRI recordings: A 7T study

https://doi.org/10.1016/j.neuroimage.2019.02.021Get rights and content

Highlights

  • Pulse artifacts (PAs) remain a challenging problem for EEG-fMRI analysis.

  • This study investigates the properties and sources of PA variability across beats.

  • PA variability is linked to head motion, respiration, and heart rate fluctuations.

  • Different correction methods differ in their ability to account for PA variability.

  • Despite its wide use, AAS cannot fully account for short-timescale PA variability.

Abstract

Electroencephalography (EEG) recordings performed in magnetic resonance imaging (MRI) scanners are affected by complex artifacts caused by heart function, often termed pulse artifacts (PAs). PAs can strongly compromise EEG data quality, and remain an open problem for EEG-fMRI. This study investigated the properties and mechanisms of PA variability across heartbeats, which has remained largely unaddressed to date, and evaluated its impact on PA correction approaches. Simultaneous EEG-fMRI was performed at 7T on healthy participants at rest or under visual stimulation, with concurrent recordings of breathing and cardiac activity. PA variability was found to contribute to EEG variance with more than 500 μV2 at 7T, which extrapolates to 92 μV2 at 3T. Clustering analyses revealed that PA variability not only is linked to variations in head position/orientation, as previously hypothesized, but also, and more importantly, to the respiratory cycle and to heart rate fluctuations. The latter mechanisms are associated to short-timescale variability (even across consecutive heartbeats), and their importance varied across EEG channels. In light of this PA variability, three PA correction techniques were compared: average artifact subtraction (AAS), optimal basis sets (OBS), and an approach based on K-means clustering. All methods allowed the recovery of visual evoked potentials from the EEG data; nonetheless, OBS and K-means tended to outperform AAS, likely due to the inability of the latter in modeling short-timescale variability. Altogether, these results offer novel insights into the dynamics and underlying mechanisms of the pulse artifact, with important consequences for its correction, relevant to most EEG-fMRI applications.

Introduction

Scalp electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) can be combined to achieve rich descriptions of the electrical and hemodynamic processes underlying brain function, non-invasively and with high spatiotemporal resolution (Mulert and Lemieux, 2010; Ullsperger and Debener, 2010; Jorge et al., 2014). Unfortunately, however, when acquired together, EEG and fMRI can interfere with each other through electromagnetic interactions that can strongly compromise data quality (Mullinger and Bowtell, 2011). Namely, the ongoing function of the heart in the static magnetic field of MRI scanners (B0) produces a major contribution to EEG degradation, hereafter termed pulse artifact (PA) – also often named “ballistocardiogram artifact”. In the spectral domain, PA contributions can span a frequency range up to 30 Hz, overlapping with an important part of the EEG spectrum. Even at moderate field strengths of 1.5T, PAs can often reach amplitudes of 50 μV (Debener et al., 2008; Neuner et al., 2013), easily surpassing most contributions from neuronal activity. At higher fields this becomes even more problematic, as PA amplitude increases approximately linearly with B0 strength (Debener et al., 2008). For these reasons, PAs are a crucial nuisance for simultaneous EEG-fMRI, and have been given considerable attention (Ullsperger and Debener, 2010).

Several mechanisms have been shown to contribute to the generation of PAs, including bulk head motion in B0 due to blood ejection from the heart, electric fields generated by ion separation in the moving blood (Hall effect), and local electrode movements due to scalp expansion (Yan et al., 2010; Mullinger et al., 2013). Unfortunately, being a consequence of heart function and vascular physiology in the MR environment, these contributions cannot be avoided at their origin, and post-acquisition PA correction methods have become ubiquitous steps in EEG data processing.

The cyclic character of cardiac function motivated early on the development of average artifact subtraction (AAS), a technique which segments the EEG signal into epochs corresponding to the cardiac cycles, estimates the artifact in each epoch as an average over several nearby epochs, and subtracts it from the signal (Allen et al., 1998). Despite its simplicity, AAS has proved effective, and remains arguably the most widely used correction approach in EEG-fMRI, along with variants building upon the same principle (Goldman et al., 2000; Sijbers et al., 2000; Ellingson et al., 2004). AAS typically employs local windowed-averaging, where each epoch is corrected by an average over a number of its closest epochs in time. This is intended to account for changes in the artifact shape and amplitude over time, presumably caused by drifts in head position and orientation (Allen et al., 1998; Moosmann et al., 2009). However, as reported by various groups (Allen et al., 1998; Debener et al., 2007; Grouiller et al., 2016) and shown in this work as well (Fig. 1), important PA residuals are often found after AAS correction, which appear to be caused by rapid variations in the artifact across epochs, and even across consecutive epochs. This type of beat-to-beat variability cannot be addressed by AAS, even with very small averaging windows (Fig. 1).

While considerable attention has been devoted to understanding the mechanisms that generate PAs, the sources of their inter-epoch variability remain, to our knowledge, largely unaddressed. This is, however, a crucial aspect for EEG analysis, given the impact that it can have on the performance of correction techniques such as the widely-used AAS approach. A deeper understanding of PA variability could prove extremely valuable to better discern the extent of artifact residuals following EEG data correction, and to guide new improvements in correction techniques.

Considering human anatomy and physiology, and their interactions with magnetic fields, a number of sources could be hypothesized to contribute to PA variability. As often discussed in the literature, changes in head position and orientation will change the projection of the EEG wire loops along B0, and can also affect the dynamics of pulsatile head motion, scalp expansion and vessel orientation in B0, leading to changes in PA amplitude and shape (Yan et al., 2010). Far less often taken into consideration, respiration may also contribute to PA variability (Allen et al., 1998). From a physiology standpoint, it is known that, even at rest, systolic blood pressure exhibits a cyclic variation that is coupled to the breathing cycle, with inspiratory periods being associated to a lower pressure compared to expiratory periods – a mechanism known as “respiratory waves in arterial pressure” (Guyton and Hall, 2006). Such pressure variations, affecting the blood ejected by the heart, could likely have an impact in all three putative PA sources, inducing beat-to-beat variability on the relatively short timescale of the breathing cycle. Additionally, it is known that respiration affects the local magnetic susceptibility, leading to cyclic fluctuations in the B0 field distribution, and can also induce small cyclic movements of the head (van Gelderen et al., 2007). While the former effect is likely too small to play a relevant role (B0 variations of only a few parts-per-million (Jorge et al., 2018)), the induced cyclic changes in head position/orientation could potentially have a measurable influence on the ongoing PAs (Gretsch et al., 2018).

Another element that may play a relevant role in PA variability is heart rate itself. Fluctuations in heart rate can cause measurable fMRI signal changes in the brain, especially near regions with cerebrospinal fluid or blood vessels (Chang et al., 2009), suggesting an influence of heart rate on vascular properties. Moreover, in EEG-fMRI, PAs can extend for at least 600 ms following each QRS event (Debener et al., 2008; Mullinger et al., 2013), and potentially even affect subsequent artifact epochs (de Munck et al., 2013; Vincent et al., 2007), suggesting that the observed PA shape may change with variations in heart rate. Another (non-causal) link may arise from the fact that sympathetic and parasympathetic mechanisms that can regulate heart rate also act upon the cardiac contractile strength (and therefore the systolic pressure) simultaneously (Guyton and Hall, 2006).

Therefore, the aim of the present study was to investigate the properties and mechanisms of pulse artifact variability in concurrent EEG-fMRI, and to evaluate its impact on various PA correction approaches. Considering the mechanisms discussed above, we hypothesized that respiration, heart rate, and head motion may be linked to PA variability. Simultaneous EEG-fMRI was performed on healthy human participants at rest or undergoing visual stimulation, with concurrent recordings of breathing and cardiac activity. The acquisitions were performed at 7T, where the PA has a larger signal-to-noise ratio than in more conventional studies. The contributions of PA variability to the recorded EEG signals were first assessed by comparison to additional datasets that do not contain PAs, including: off-scanner human EEG, expected to contain only real EEG activity and EEG-specific artifacts (e.g. eye-blinks, muscle activity), and phantom EEG-fMRI, expected to contain only MR environment-related artifacts (e.g. He coldheads, which can be important at higher fields (Mullinger et al., 2008), and gradient artifact residuals). In a subsequent part, the human EEG-fMRI resting-state data were analyzed with clustering techniques, to determine potential relationships between the variability across different PA epochs and the ongoing respiratory amplitude, cardiac period, head position and orientation. Potential confounds, such as gradient artifact residuals or ongoing EEG activity, were also considered. PA correction was then studied on the visual stimulation data with three distinct approaches, each with specific strengths and limitations in accounting for PA variability: AAS, optimal basis sets (OBS), and a K-means clustering-based method.

Section snippets

Materials and methods

This study was approved by the local ethics committee (CER-VD), and involved the participation of 12 healthy volunteers (23 ± 3 years old, 6 male/6 female), who provided written informed consent. All collected data are available upon request, in coded form to preserve anonymity.

Results

Across the 12 participants of this study, an average of 461 ± 1 PA epochs (range: 409–529) were recorded per subject during the 8-min EEG-fMRI resting-state run, with an average cardiac period of 1.06 ± 0.03 s (0.92–1.18 s).

Discussion

This study investigated the variability of EEG pulse artifacts across heartbeats, in simultaneous EEG-fMRI. While previous reports have pointed to the existence of PA variability, presumably caused by changes in head position and orientation throughout the recordings (Allen et al., 1998; Bonmassar et al., 2002), the present study is, to our knowledge, the first to objectively assess the magnitude and properties of PA variability, and to investigate its underlying mechanisms, including, but not

Conclusion

We conclude that the variability of pulse artifacts across heartbeats has an important impact on EEG data quality acquired in the MRI environment. PA variability is linked to changes in head position and orientation, as previously hypothesized, but also, and more importantly, to respiration and to heart rate. The latter mechanisms are associated with short-timescale variability that cannot be fully captured by PA correction techniques based on windowed averaging, such as AAS, which tended to be

Disclosure/conflicts of interest

The authors declare no conflicts of interest.

Acknowledgments

This work was supported by Centre d'Imagerie BioMédicale (CIBM) of the UNIL, UNIGE, HUG, CHUV, EPFL and the Leenaards and Jeantet Foundations.

References (48)

  • M.L. Ellingson et al.

    Ballistocardiogram artifact reduction in the simultaneous acquisition of auditory ERPS and fMRI

    Neuroimage

    (2004)
  • R.I. Goldman et al.

    Acquiring simultaneous EEG and functional MRI

    Clin. Neurophysiol.

    (2000)
  • S.I. Goncalves et al.

    Artifact removal in co-registered EEG/fMRI by selective average subtraction

    Clin. Neurophysiol.

    (2007)
  • F. Grouiller et al.

    A comparative study of different artefact removal algorithms for EEG signals acquired during functional MRI

    Neuroimage

    (2007)
  • M. Jenkinson et al.

    Improved optimization for the robust and accurate linear registration and motion correction of brain images

    Neuroimage

    (2002)
  • J. Jorge et al.

    Towards high-quality simultaneous EEG-fMRI at 7 T: Detection and reduction of EEG artifacts due to head motion

    Neuroimage

    (2015)
  • J. Jorge et al.

    Simultaneous EEG-fMRI at ultra-high field: artifact prevention and safety assessment

    Neuroimage

    (2015)
  • J. Jorge et al.

    EEG-fMRI integration for the study of human brain function

    Neuroimage

    (2014)
  • P. LeVan et al.

    Ballistocardiographic artifact removal from simultaneous EEG-fMRI using an optical motion-tracking system

    Neuroimage

    (2013)
  • D. Mantini et al.

    Complete artifact removal for EEG recorded during continuous fMRI using independent component analysis

    Neuroimage

    (2007)
  • M. Moosmann et al.

    Realignment parameter-informed artefact correction for simultaneous EEG-fMRI recordings

    Neuroimage

    (2009)
  • K. Mullinger et al.

    Exploring the feasibility of simultaneous electroencephalography/functional magnetic resonance imaging at 7 T

    Magn. Reson. Imaging

    (2008)
  • K.J. Mullinger et al.

    Identifying the sources of the pulse artefact in EEG recordings made inside an MR scanner

    Neuroimage

    (2013)
  • I. Neuner et al.

    EEG acquisition in ultra-high static magnetic fields up to 9.4 T

    Neuroimage

    (2013)
  • Cited by (0)

    View full text