Classic EEG (ERPs)/ Advanced EEG Quentin Noirhomme
Outline Origins of MEEG Event related potentials Time frequency decomposition i Source reconstruction
Before to start EEGlab Fieldtrip (included in spm)
Part I: Origins EEG Discovered by Hans Berger in 1924 Non invasive measure of electrical brain activity
Origins: MEG 1968
Origins Baillet et al., IEEE Sig. Proc. Mag., 2001
Origins: Potentials
Origins Baillet et al., IEEE Sig. Proc. Mag., 2001
M/EEG vs. fmri
Raw EEG
Fp2 T4 EEG in coma Burst Suppression Alpha coma Isoelectric T4 02 Fp2 C4 C4 02 Fp1 T3 T3 01 Fp1 C3 C3 01 50 µv 50 µv 20 µv 20 µv 1 s 1 s 1 s Thömke et al. BMC Neurology 2005 5:14 doi:10.1186/1471 2377 5 14
EEG in sleep http\\:www.benbest.com
EEG Rhythms Gamma : > 30 Hz http://members.arstechnica.com/x/albino_eatpod/specific eeg states.gif
Burst EEG events Spikes http://members.arstechnica.com/x/albino_eatpod/specific eeg states.gif
Part II: Event Related potentials Wolpaw et al., 2000
Averaging Adapted from Tallon Baudry and Bertrand, 1999 Average potential (across trials/ subjects) relative to some specific event in time
Preprocessing 1. Filtering 2. Segmentation 3. Artifact rejection 4. Averaging 5. Baseline removal
Filtering Why filter? EEG consists of a signal plus noise Some of the noise is sufficiently different in frequency content from the signal that it can be suppressed simply by attenuating ti different frequencies, thus making the signal more visible Non neural physiological activity (skin/sweat potentials) Noise from electrical outlets Highpass filter to remove drift due to sweating, Notch filter to remove the line noise (50 60Hz) Low pass filter (often 30Hz for ERP)
Segmentation
Artifacts
Artifacts http://www.bci2000.org
Artifacts http://www.bci2000.org
Artifacts http://www.bci2000.org
Artifacts http://www.bci2000.org
Artifact rejection Visual inspection of the data Thresholding (e.g., everything above 100µV) Statistical i method Independent component analysis good for blinks and other visual artifacts Help if you have EOG and EMG channels Do not trust automatic methods
Averaging
Averaging Assumes that only the EEG noise osevaries from trial to trial But amplitude and latency will vary
Averaging: effects of variance L t i ti b Latency variation can be a significant problem
Averaging Assumes that only the EEG noise osevaries from trial to trial But amplitude and latency will vary S/N ratio increases as a function of the square root of the number of trials. It s always better to try to decrease sources of noise than to increase thenumberof trials.
Baseline correction Remove the mean of the recorded baseline (e.g., 200 ms to 0 ms) Variation in baseline duration can induce change in potential amplitude Individually id for each electrode SPM does it automatically while segemting the data
Part III: Time frequency decomposition Adapted from Tallon Baudry and Bertrand, 1999
Evoked frequency Adapted from Tallon Baudry and Bertrand, 1999
Induced frequency decomposition Adapted from Tallon Baudry and Bertrand, 1999
Induced frequency decomposition Adapted from Tallon Baudry and Bertrand, 1999
Time frequency decomposition Adapted from Tallon Baudry and Bertrand, 1999
Continuous Morlet wavelet http://amouraux.webnode.com/.
Analysis Grand mean > > Average across subject Convert ERP or TF decomposition into images => first/second level lanalysis Source reconstruction => first/second level analysis
1 st Level Analysis select periods or time points in peri stimulus time Choice made a priori. sum over all time points
Part IV: Source reconstruction From www.imt.uni luebeck.de, 2008
Source reconstruction 1. Forward Model 2. Inverse reconstruction
Forward modeling Electromagnetic head model Reconstruct electrode signals from electrical current in the head
Head model Spherical approximation Realistic head model Boundary element method Finite element method
SPM head model Compute transformation T Individual MRI Templates Apply inverse transformation T 1 Individual mesh BEM mesh
Head model Electrode locations Registration Landmark kbased Surface matching Leadfield fiducials fiducials Rigid transformation (R,t) Individual sensor space Individual MRI space
Inverse approaches Dipole Distributed dipoles Least square or Beamforming More unknowns than data
Distributed approach Y = KJ+ E No unique solution! Pi Priors: min( Y KJ 2 + λf(j) ) minimum overall activity Location Smoothness Bayesian model comparison
References Sylvain Baillet s presentation at HBM 2008 SPM for dummies 0000 2008 presentations http://www.bci2000.org Baillet et al., IEEE Sig. Proc. Mag., 2001 Mtt Mattout, tphilli Phillips & Fit Friston (2005) SPM course http://www.fil.ion.ucl.ac.uk/spm/course/slide s05/ppt/meeg_inv.ppt t SPM manual