Center for Neuroscience UNIVERSITY OF CALIFORNIA AT DAVIS ERPs in Cognitive Neuroscience Charan Ranganath Center for Neuroscience and Dept of Psychology, UC Davis
EEG and MEG Neuronal activity generates extracellular electrical and magnetic fields that are measured by EEG and MEG, respectively EEG system MEG system
What is an Event Related Potential (ERP)?
What will/won t produce an ERP NO YES! Example from Kutas & Dale (1997), in Cognitive Neuroscience, M. D. Rugg (ed.),
Sources of EEG and MEG signals Radial Source Primary currents Tangential Source Courtesy Matti Hämäläinen
Characteristics of EEG and MEG EEG Sensitive to radial and tangential sources Electrical fields are distorted by skull, scalp MEG Generally sensitive to tangential sources only Magnetic fields not distorted by skull Source localization is more tractable for MEG But that does not mean that MEG is better than EEG
EEG and FMRI EEG is related to synchronized synaptic potentials in cortical pyramidal cells BOLD signal (FMRI) is largely driven by metabolic demands related to synaptic activity So, if condition A elicits increased synchronous synaptic potentials relative to condition B, there may be comparable neocortical sources for ERPs and FMRI signal in A-B contrast If ERP is related to phase-reset of ongoing oscillation then ERPs might be seen with no BOLD If synaptic activity changes are not synchronous then BOLD might be seen with no ERPs
Typical Visual ERP Waveform
WHAT IS AN ERP COMPONENT? A fuzzy concept with varying definitional criteria: Timing of positive/negative deflections or peaks Scalp Topography Functional characteristics Neural generators (sources)
Example: P300 Sound- Elicited ERPs Light- Elicited ERPs P300 P300 Scenario: Cue stimulus indicating whether click or flash was likely Delay of 3-5 seconds: Subject guesses whether stimulus will be click or flash Click or flash occurs Sutton, S., Braren, M., Zubin, J., & John, E. R. (1965) AAAS
Different P300 components: P3a vs P3b Timing of positive/negative deflections or peaks P3a: Positive peak ~200-300ms P3b: Positive peak ~300-400ms
Different P300 components: P3a vs P3b Scalp Topography P3a: Fronto-central topography P3b: Parietal topography
Different P300 components: P3a vs P3b Neural Generators P3a: Prefrontal Cortex, Hippocampus, Temporoparietal P3b: Temporoparietal
WHAT IS AN ERP COMPONENT? A fuzzy concept with varying definitional criteria: Task-based definitions Error-related negativity NOTE NEGATIVE PLOTTED UP! Gehring et al., Psych. Science (1993) Cohen & Ranganath, J. Neurosci. (2007)
WHAT IS AN ERP COMPONENT? A fuzzy concept with varying definitional criteria: Task-based definitions Error-related negativity Contingent Negative Variation Lateralized readiness potential NOTE NEGATIVE PLOTTED UP! Gehring et al., Psych. Science (1993) Cohen & Ranganath, J. Neurosci. (2007)
WHAT IS AN ERP COMPONENT? Data-driven approaches to component identification Principal or Independent Component Analysis Makeig et al. (1997) PNAS
Overlapping component problem ERP with one dorsomedial source Example from Kutas & Dale (1997), in Cognitive Neuroscience, M. D. Rugg (ed.),
Overlapping component problem Two simultanously active sources Example from Kutas & Dale (1997), in Cognitive Neuroscience, M. D. Rugg (ed.),
Overlapping Component Problem Which plot is from Source M and which is from Source L + R? Example from Kutas & Dale (1997), in Cognitive Neuroscience, M. D. Rugg (ed.),
Problems with the component approach Overlapping component problem Many important ERP modulations occur after 300ms post-stimulus By this time there is massively parallel processing in the brain A task manipulation might not affect the component of interest but instead result in the generation of a new, temporally-overlapping component
An MEG Study of Word Repetition Dale et al., 2000 repeated words vs. novel words
Experimental design tips for good ERP studies It is fine to capitalize on a known component, but try not to have your experimental interpretation hinge on a particular component Because of overlapping components, reverse inferences may be invalid Solution: Design experiments so critical result is a difference between two or more conditions For example, cognitive subtraction or parametric designs, analagous to imaging studies But, beware of known components that can confound your data Example: P300
Experimental design tips for good ERP studies Eye movements and blinks create substantial EOG artifacts in EEG Artifacts can be corrected but better to avoid problems Solutions: Keep visual stimuli simple, foveal, and present for as little time as possible Include blink breaks in experiment Insert at least 2s of preparatory cue between blink break and trials Try to avoid contact lens wearers
Experimental design tips for good ERP studies SNR is a function of sqrt(# of trials) Doubling # trials increases SNR by 41% [sqrt(2)=1.41] Quadrupling # trials doubles SNR [sqrt(4)=2] Get LOTS of trials Slide courtesy S.J. Luck
Data acquisition
Electrode Impedance Measured in Ohms Reduced by using conductive gel and abrading scalp For scalp electrodes <5 Ohms = low impedance Newer systems (EGI, Bio-Semi, etc.) have high input impedance so you don t need to abrade scalp to get high quality recordings But with high electrode impedance, large, low frequency skin potentials can occur Cephalic skin potentials are large, slow potentials that occur when the autonomic nerves and sweat glands in the skin are activated by heat or arousal (Picton & Hillyard, 1972). They are most prominently observed in the forehead, temples, neck, and mastoid regions.
Luck & Kappenman (in press) High impedance does not globally attenuate ERP component amplitude Large, sudden changes in EEG were seen with high impedance recordings, esp. at high temperature. Changes primarily in frequencies below 10Hz Removing voltage fluctuations >100 microvolts helps attenuate artifacts.
Electrode Impedance and EEG Frequency Content Slide courtesy S.J. Luck
High-Impedance & Noise Direct comparison of high & low Z in Biosemi system Oddball paradigm (N=12); cool/dry vs. warm/humid Look at significance of rare vs. frequent P3 difference Slide courtesy S.J. Luck
Electrode cap and locations Original International 10/20 System
1994 Revised 10/20 System
Do I need >128 channels? Pros: More accurate topographic, scalp current density maps Necessary for source localization Makes average reference useful Not difficult with contemporary high impedance systems Cons: Takes longer to apply electrodes Subject fatigue more likely Increased likelihood of bridging More electrodes -> > More chances for problems Data overload!
Tips on getting good ERP data (an incomplete list) Keep chamber cool Reduces skin potentials Note you needn t use a shielded chamber! Don t use too much gel Reduces likelihood of electrode bridging Keep participants comfortable and relaxed Reduces muscle (EMG) artifact Use extra care in ensuring good data from reference electrode Try to avoid EKG artifact in mastoids
Data processing and analysis 1. Resampling 2. Re-referencing 3. Filtering 4. Artifact rejection/correction 5. Binning/Averaging 6. Normalization
Data processing and analysis 1. Resampling 2. Re-referencing 3. Filtering 4. Artifact rejection/correction 5. Binning/Averaging 6. Normalization
For your Reference EEG is a relative measure Data is usually collected relative to a reference electrode No matter what reference is used during recording, you can algebraically re-reference the data offline Choice of reference site will alter observed scalp topography
Reference = Left Mastoid Reference= Average of Fz, Cz, Pz Reference = Average of Fz, Cz, Pz, O1/O2, and T5/T6 Slide courtesy S.J. Luck
For your Reference Referencing strategies: 1. Relative to a electrically dead site Averaged/linked mastoid electrodes Nose-tip, earlobe If you use this approach make sure reference site is not close to regions that may generate your ERP effect! 2. Averaged reference Note that by definition, each positive effect will be accompanied by negative effects somewhere else This method is closest to reference free measures if you have a lot of electrodes
Scalp Current Density Another option is to convert the data into current density This reflects the current flowing outward at each point of the scalp; Reference-independent Calculated as the 2nd derivative over space Emphasizes superficial sources Estimates are poor at edges of electrode array Requires large # of electrodes and high SNR (lots of trials) Voltage Current Density Courtesy S.J. Luck
Filtering your data Filtering sacrifices the high temporal precision of ERPs, so avoid excessive offline (digital) filtering ERPs look ugly with high-frequency noise, but that s reality Static, high-frequency noise will generally average out if you have enough trials Even short-duration components have some lowfrequency contributions, so high-pass filtering may distort timing and amplitude of these components Necessary to high pass filter data if you are going to do independent component analysis (<.25 Hz)
Slide courtesy S.J. Luck Artifacts: Blinks
Artifacts: Eye movements Active: HEOG-L Reference: HEOG-R Slide courtesy S.J. Luck
Artifacts: C.R.A.P. (Commonly Recorded Artifactual Potentials) Slide and acronym courtesy S.J. Luck
Dealing with artifacts Artifact rejection Manual (Visual Inspection) Automated Fixed peak-to-peak amplitude thresholds Slope Difference between frontal and vertical EOG electrode
Artifact Correction: ICA Artifact Correction: ICA http://www.sccn.ucsd.edu/~scott/tutorial/icatutorial8.html
Averaging and Baselines Baseline correction Usually done relative to average of activity during pre-stimulus period Duration of baseline is usually 100-200ms Watch out for pre-stimulus activity Artifacts, Alpha Oscillations Meaningful pre-stimulus activity may confound baseline-corrected ERPs Urbach & Kutas (2006)
You have ERPs, now what? Compute grand average Average of each subject s average Examine difference waves Sometimes timing of differences differs from peaks in single-condition waveforms Johnson et al. (1998)
You have ERPs, now what? Closely examine maps of scalp topography Look for changes in topography over time and across conditions In general, indicates change in process/brain networks that are recruited Example from Kutas & Dale (1997), in Cognitive Neuroscience, M. D. Rugg (ed.),
You have ERPs, now what? Closely examine maps of scalp topography Look for changes in topography over time and across conditions In general, indicates change in process/brain networks that are recruited remember faces forget faces Paller, Bozic, Ranganath et al. (1999) Brain Res.
You have ERPs, now what? Closely examine maps of scalp topography Look for changes in topography over time and across conditions In general, indicates change in process/brain networks that are recruited Compare surface potential maps with scalp current density (SCD) maps Acts as a spatial filter to remove deep sources
Measurement and Analysis Different approaches, different assumptions 1. Peak amplitude and latency measures Used for ERP components Sometimes done relative to pre-stimulus baseline, sometimes relative to a preceding peak Problems: Peak Component: Peaks are NOT special May be excessively sensitive to noise Carries strong assumptions about components
Measurement and Analysis Different approaches, different assumptions 2. Mean amplitude measures Need to choose windows carefully Can choose windows that surround a component Can choose arbitrary measurement windows Example: consecutive 100ms windows Trade-off: wider windows can increase SNR but not if you are including time before or after effect of interest
Thanks for your attention!