Proceedings of the meeting - Department of Physics - University of ...

Recent advances in EEG/fMRI integrationStefan Debener 1,21 MRC Institute of Hearing Research, Royal South Hants Hospital, Southampton SO14 OYG, UK2 Clinical Neurosciences, School of Medicine, University of Southampton, Southampton, UKI2IntroductionElectroencephalography (EEG) and functional magnetic resonanceimaging (fMRI) provide complimentary information regarding thetemporal and spatial resolution of brain activity. Accordingly, theintegration of both signals holds great promise for a betterunderstanding of how cognitive processes are implemented in distinctcortical areas. Following an overview about the virtues and pitfalls ofsimultaneous and separate recordings, it will be described howartifacts unique to inside scanner EEG recordings can be removed. Itwill be shown that, following scanner artifact correction, signals ofsufficient quality can be obtained from inside scanner EEG recordings(1). Importantly, this enables the application of advanced single-trialoriented EEG analysis techniques such as independent componentanalysis (ICA, e.g., 2). A recent EEG/fMRI study on cognitiveperformance monitoring will be presented showing that thecombination of simultaneous recordings and ICA enables to study thedirect trial-by-trial coupling between EEG and fMRI data (3). Thebenefits of this new integration approach for the study of cognitivefunction will be discussed.Mining event-related brain dynamicsUsually only the evoked fraction of the EEG signal, the time-domainaveraged event-related potential (ERP), is considered and all trial-bytrialfluctuations are regarded as noise. However, trial-by-trialfluctuations of event-related EEG signals are meaningful, since theyhave been shown to be specifically correlated with the BOLD signal(4). From an information-based point of view, it is reasonable toconsider event-related EEG signals within a multi-dimensional signalspace (5). It is proposed that the consideration of EEG signals beyondthe ERP can help to investigate the coupling betweenelectrophysiological and hemodynamic properties of brain activity.Separate versus simultaneous recordingsPrevious EEG/fMRI integration attempts often focused on separatelyrecorded signals, and combined the results of both measures ratherthan the signals directly. However, separate EEG/fMRI recordingprotocols are not well suited for the investigation of a possible directcoupling between hemodynamic and electrophysiological measures ofbrain activity. More recent advances in EEG hardware and softwaredevelopments have made it feasible to simultaneously record EEG andfMRI data. Only simultaneous recordings guarantee that the identicalsensory, mental, and behavioral conditions are being studied.Moreover, simultaneous recording protocols allow to jointly exploitthe trial-by-trial fluctuations of both measures, which can revealimportant insights into the dynamic properties of cognitive function.Quality of EEG signals recorded inside the MRIThe quality of EEG data recorded simultaneously with fMRI ismassively compromised by gradient artefacts and ballistocardiogramartefacts (Figure 1). Gradient artefacts can cause amplitudes beingorders of magnitudes larger than the EEG signal, but are relativelyinvariant over time, and therefore can be subtracted out (Figure 1 B).However, the major challenge is the ballistocardiogram (BCG)artefact, which originates from the pulsatile movement of blood andthe pulsatile movement of EEG electrodes placed adjacent to largerblood vessels (Figure 1 B-D). Both channel-wise and spatial filtertechniques have been proposed for BCG-correction. Results of asystematic comparison of three different BCG-correction approaches(1) will be presented. It will be shown that a multivariate combinedchannel-wise and spatial filter approach based on ICA is superior toalternative techniques. Following appropriate BCG-correction, EEGsignals of reasonable quality can be obtained.Direct coupling between event-related EEG andfMRI in a cognitive taskGoal-directed behavior requires the continuous monitoring and dynamicadjustment of ongoing actions. We simultaneously recorded 32channel EEG and fMRI data from subjects performing an Eriksenflanker task (3). By applying ICA to the EEG signals, we found thesingle-trial error-related negativity of the EEG to be systematicallyrelated to behavior in the subsequent trial. This result reflects theimmediate behavioral adjustment of a cognitive performance monitoringsystem. Moreover, the trial-by-trial EEG measure of performancemonitoring predicted the fMRI activity in the rostral cingulate zone, abrain region known to play a key role in processing of response errors.Importantly, this result was more specific when compared to a conventionalfMRI analysis, which supports the conclusion that temporalinformation, provided by the EEG, can improve the processing offMRI signals. Furthermore, the study shows that it is possible to investigatethe dynamic coupling of non-invasively recorded EEG andfMRI data. This new approach, EEG-informed fMRI analysis, providesa powerful way for the study of brain function related to cognitiveprocessing.Figure 1: Example 10-sec traces from inside scanner EEG (A-C) andEKG (D) recordings. EEG data from six representative channels areshown before (A) and after correction for MR gradient (B) andballistocardiogram (C) artefacts.References(1) Debener S et al. submitted (2006)(2) Debener S et al. Cog Brain Res. 22, 309-321 (2005)(3) Debener S et al. J Neurosci. 25, 11730-11737 (2005)(4) Debener S et al. Neuroimage 31, S31 (2006)(5) Makeig S et al. Trends in Cog Sci 8, 204-210 (2004)