ICA-based procedures for removing ballistocardiogram artifacts from EEG data acquired in the MRI scanner
Introduction
Combining information from EEG and fMRI holds great promise for examining the spatial and temporal dynamics of sensory and cognitive processes underlying brain function (Ahlfors et al., 1999, Babiloni et al., 2000a, Babiloni et al., 2000b, Dale et al., 2000, George et al., 1995, Menon et al., 1997, Woldorff et al., 1999). Concurrent acquisition of EEG and fMRI has proven to be challenging for a number of reasons including safety and data quality (Goldman et al., 2000). In the past 5 years, several brain imaging centers have developed and refined techniques for simultaneous acquisition of EEG and fMRI data (Bonmassar et al., 2001, Krakow et al., 2000, Lemieux et al., 2001, Salek-Haddadi et al., 2002) and have used them to detect EEG spikes, characterize resting state EEG and fMRI, and study event-related potentials (ERPs) (Bonmassar et al., 1999, Liebenthal et al., 2003). Common to all these studies is the problem of removing BCG artifacts which contaminate EEG data in the scanner (Allen et al., 1998, Bonmassar et al., 2002, Sijbers et al., 2000). In this paper, we demonstrate the application of ICA-based procedures to remove BCG artifacts from simultaneous EEG data recorded in the MRI scanner. We show that ICA-based procedures can significantly reduce the spectral power associated with BCG artifacts and that the performance of these procedures is superior to BCG artifact removal techniques that rely on averaged artifact subtraction (Allen et al., 1998, Goldman et al., 2000).
BCG artifacts are a consequence of electromotive force produced on the EEG electrodes due to small head movements, such as those caused by cardiac pulsation, inside the scanner magnetic field. As pointed by Sijbers et al. (2000), there are three major sources of BCG artifacts: (1) small but firm movement of the electrodes and the scalp due to expansion and contraction of scalp arteries between systolic and diastolic phase, (2) fluctuation of the Hall-voltage due to the pulsatile changes of the blood in the arteries, and (3) small cardiac related movements of the body. The cardiac pulse generates artifacts with amplitudes considerably larger than EEG signal fluctuations. It is therefore important to develop methods to identify and remove these artifacts in a robust manner.
Most approaches for eliminating BCG artifacts to date have focused on either (1) Averaged Artifact Subtraction (AAS), in which a BCG artifact template is estimated by averaging over the intervals of EEG signal that are corrupted by the artifact and subsequent subtraction of the template from the corrupted segments to obtain clean signal (Allen et al., 1998) or (2) Adaptive filtering techniques, which make use of correlations between a reference ECG channel and the EEG channels to estimate the contribution of BCG artifact in the EEG signals which is then subtracted to give clean signals (Bonmassar et al., 2002).
The AAS procedure is the most commonly employed method for removing the BCG artifact from EEG data (Allen et al., 1998). In this procedure, the QRS peaks in the ECG signal are first detected and then EEG activity time-locked to these peaks is averaged to give an estimate of the pulse artifact. The average artifact is then subtracted from the EEG. Goldman et al. (2000) have used a method that is conceptually similar to AAS procedure, but it differs in the weights applied to data segments prior to averaging. These weights vary inversely with the temporal displacement from the current sample to compensate for the slow changes in the BCG artifact. Along similar lines, Sijbers et al. (2000) have used QRS onset detection for creating a template of the BCG artifact based on adaptive filtering. They point out that simple averaging would not lead to a satisfactory template, as the ECG is not a stationary signal and hence the rate and duration of BCG artifacts may vary over time. So in their approach, median filtering was performed for obtaining an artifact template because it adapts to changes in ECG signals over time. Bonmassar et al. (2002) used motion information recorded from a piezoelectric sensor placed on the temporal artery to estimate the motion artifact noise (mostly BCG), followed by adaptive filtering to subtract the artifact.
Automated AAS procedures are not amenable for removing BCG artifacts in the interleaved scanning scenario which is described in Materials and methods. In such a case, the duration of BCG artifacts may vary from trial to trial since some of the BCG artifacts lie partially in scanning and non-scanning intervals. These cannot be removed satisfactorily by AAS as it relies on the estimation of artifact template for the entire duration of the artifact. One apparent problem with the adaptive filtering approach is that it assumes that the reference signal contains only the activity of the source of noise and not other neural signals. This assumption may not be true if the reference signal is also acquired on or near the scalp because in that case, it will have contributions from other neural generators and the correlations between the reference signal and the EEG signal will give an erroneous estimate of noise. Using such methods might lead to removal of useful neurophysiological signals as well. Therefore, new procedures are required which can overcome the problem of variability in ECG pulse rate and duration, the need for an artifact template, and the need for a reference ECG channel. An ICA-based procedure provides a potential approach to circumvent these problems as it makes no assumptions about the morphology (rate or duration) of the mixing signals and does not necessitate the use of a reference signal for extracting BCG artifacts.
ICA can be used to recover independent sources from a set of simultaneously acquired signals that result from a linear mixing of the source signals (Comon, 1994). The ICA algorithm makes no assumption about the mixing process except that it is linear. A good intuitive mathematical formulation of ICA is given in Jung et al., 2001a, Jung et al., 2001b.
A number of procedures have been developed in recent years to isolate the source signals (Hyvarinen et al., 2001); here, we identify independent components (ICs) using the infomax approach (Bell and Sejnowski, 1995, Lee et al., 1999).
ICA has been used successfully by Tong et al. (2001) to remove ECG interference from EEG recordings in small animals. Applied to human EEG data, ICA has been used to separate complex multichannel data into spatially fixed and temporally independent components without requiring detailed models of either the dynamics or the spatial structure of the separated components (Jung et al., 2001a, Jung et al., 2001b). In recent years, ICA has become increasingly popular for artifact removal and characterizing distinct or overlapping brain or extra-brain sources of activations. It has been employed for analysis of single-trial event related potentials, for removal of blinks, eye movements, temporal muscle activity, and electrode artifacts (Iriarte et al., 2003, Jung et al., 2000a, Jung et al., 2000b, Jung et al., 2001a, Jung et al., 2001b).
Here, we apply ICA-based procedures for isolating and removing BCG artifacts from human EEG data acquired in the scanner. One weakness of previous studies in the field is that none of them have provided a quantitative comparison between procedures commonly used for removing BCG artifacts. In the absence of such a comparison, it is very difficult to assess the relative strengths of various approaches. Here, we provide quantitative evidence that our ICA-based procedures performs better than the more standard AAS-based procedures.
Section snippets
Subjects
Five right-handed healthy volunteers (4 males and 1 female, ages 20–24 years) participated in this study. All study protocols were approved by the human subjects committee at Stanford University School of Medicine, and subjects provided written informed consent prior to participation in the study.
EEG-fMRI acquisition
The spiral-in/out fMRI sequence was adapted for interleaved EEG acquisition using a clustered procedure comprising 2 s of EEG acquisition without MRI scanning, followed by 2 s of fMRI data acquisition (
EEG data before and after removal of BCG artifacts using ICA
We first present an example of the application of ICA for BCG artifact removal based on data from a representative subject (subject 1). As shown in Fig. 2A, the BCG artifacts are prominent in all the epochs and all electrodes. The artifacts are clearly time-locked to the observable peaks in the ECG channel. Fig. 2B shows the same EEG data after removal of BCG artifacts using ICA. Although the ICA analysis was carried out excluding the ECG channels, the BCG artifact has been effectively removed
Discussion
In this report, we have described new ICA-based procedures for removing BCG artifacts from EEG recorded inside an MRI scanner. ICA decomposition allowed us to identify and remove BCG artifacts in a reliable manner. ICA procedures were efficient in reducing spectral power in the fundamental ECG frequency and its higher harmonics at all the scalp electrodes. Detailed quantitative comparisons showed that ICA-based procedures performed significantly better than more standard techniques that involve
Acknowledgments
This research was supported by a grant from the Bio-X interdisciplinary initiative program at Stanford University and by a grant from the NIH (HD40761) and grant (NIH RRP4109784). We would like to thank A. Delorme and S. Makeig of Swartz Center for Computational Neuroscience, UCSD for making the EEGLAB software available for analysis.
References (36)
- et al.
Identification of EEG events in the MR scanner: the problem of pulse artifact and a method for its subtraction
NeuroImage
(1998) - et al.
Motion and ballistocardiogram artifact removal for interleaved recording of EEG and EPs during MRI
NeuroImage
(2002) Independent component analysis: a new concept ?
Signal Process., Elsevier
(1994)- et al.
Dynamic statistical parametric mapping: combining fMRI and MEG for high-resolution imaging of cortical activity
Neuron
(2000) - et al.
EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis
J. Neurosci. Methods
(2004) - et al.
Acquiring simultaneous EEG and functional MRI
Clin. Neurophysiol.
(2000) - et al.
Removal of eye activity artifacts from visual event-related potentials in normal and clinical subjects
Clin. Neurophysiol.
(2000) - et al.
Isolation of epileptiform discharges from unaveraged EEG by independent component analysis
Clin. Neurophysiol.
(1999) - et al.
Event-related fMRI with simultaneous and continuous EEG: description of the method and initial case report
NeuroImage
(2001) - et al.
Simultaneous ERP and fMRI of the auditory cortex in a passive oddball paradigm
NeuroImage
(2003)
Simultaneous EEG-correlated ictal fMRI
NeuroImage
Reduction of ECG and gradient related artifacts in simultaneously recorded human EEG/MRI data
Magn. Reson. Imaging
Removal of ECG interference from the EEG recordings in small animals using independent component analysis
J. Neurosci. Methods
Spatiotemporal activity of a cortical network for processing visual motion revealed by MEG and fMRI
J. Neurophysiol.
Integration of high resolution EEG and functional magnetic resonance in the study of human movement-related potentials
Methods Inf. Med.
Multimodal integration of high resolution EEG, MEG and functional magnetic resonance data
Int. J. Bioelectromagn.
An information-maximization approach to blind separation and blind deconvolution
Neural Comput.
Visual evoked potential (VEP) measured by simultaneous 64-channel EEG and 3T fMRI
NeuroReport
Cited by (238)
While you were sleeping: Evidence for high-level executive processing of an auditory narrative during sleep
2022, Consciousness and CognitionBallistocardiogram artifact removal in simultaneous EEG-fMRI using generative adversarial network
2022, Journal of Neuroscience MethodsCellular Classes in the Human Brain Revealed In Vivo by Heartbeat-Related Modulation of the Extracellular Action Potential Waveform
2020, Cell ReportsCitation Excerpt :Another possibility in both acute and chronic recordings is to take advantage of the motion produced through inherent physiological mechanisms, e.g., pulsatile motion of the brain caused by rhythmic changes in intracranial pressure during the heartbeat. While typically perceived as an unwanted effect that compromises data collection (Chen et al., 2012; Fee, 2000; Srivastava et al., 2005), here we propose that periodic, heartbeat-related motion improves the spatial sampling of the extracellular space. Just as high-density recordings improve unit classification compared to microwire recordings, this periodic sampling offers insights into the spatiotemporal variability of EAP features, which, in turn, can be linked to distinct, cell-class-dependent properties.
The design and implementation of multi-character classification scheme based on EEG signals of visual imagery
2024, Cognitive NeurodynamicsFunctional differences in cerebral activation between slow wave-coupled and uncoupled sleep spindles
2023, Frontiers in Neuroscience