Depth/surface relationships: Confronting noninvasive measures to intracerebral EEG Christian G Bénar Institut de Neurosciences des Systèmes; INSERM, Aix-Marseille Université christian.benar@univ-amu.fr OHBM2016 - Electromagnetical Neuroimaging course Sun, June 26th
EEG and MEG are surface recordings! MEG EEG - M/EEG are mainly sensitive to the synchronous activity of pyramidal cells in neocortex - Little influence of skull conductivity on MEG - Purely radial sources not visible in MEG
Some open questions (1) What is the extent of cortex that is necessary for recording activity on MEG/EEG? Corrolary: impact of averaging (SNR increases ~ sqrt(n) ) Grova et al Neuroimage 2006
Some open questions (2) What is the influence of depth and architectony on detectability? (e.g. hippocampus, amygdala) hippocampus amygdala Do we see these deep structures or neocortex immediately next to them?? Niedermeyer and Lopes da Silva
Some open questions (3) To what extent can EEG/MEG record high frequency activity (gamma, ripples, fast ripples) Yuval-Greenberg et al 2008 Is gamma always cortical or does it originate from muscular artefacts? Whitham et al 2007, 2008
How accurate is the inverse problem? The inverse problem is ill posed It requires mathematical hypotheses that may not have physiological grounding There are model imprecision (skull conductivity, anisotropy ) Up to 10000 sources can be computed, but how many can be accurately estimated?? Baillet et al 2001
A strategy to answer these issues: Intracranial EEG Used during presurgical evaluation of epilepsy, as a second step after noninvasive methods Allow defining the extent of cortex to be resected surgically Provides a formidable opportunity for validating non-invasive methods and for better understanding depth/surface relationships Skirrow et al Brain 2014
INTRACRANIAL RECORDINGS
Subdural grids Grid placed at the surface of the cortex (Tao et al 2005)
Stereotaxic EEG (SEEG) Electrodes places directly within the brain Bancaud and Talairach 1962 Trebuchon da Fonseca et al 2009
Example of SEEG «buildup» Schwartz et al 2011
DETECTABILITY AS A FUNCTION OF SOURCE EXTENT/ORIENTATION
Cooper et al 1965 wet skull polythene V 6cm2 (1 inch 2 ) are necessary to measure a potential at the surface
Cooper revisited Tao et al 2005: 10 cm2 for detectability in EEG Oishi et al 2002: 3 cm2 in MEG NB: raw signals; averaging increases signal to noise ratio by a factor of ~sqrt(n)
Effect of source orientation? Sleep spindle visible in: MEG +EEG EEG only MEG only Manshanden et al 2002 MEG and EEG are complementary!
Gavaret et al 2014 Case study: Bilateral occipital sources SEEG MEG reveals bilateral sources Sources are confirmed by SEEG
VALIDATION OF SOURCE LOCALIZATION IN EPILEPSY
a) Localization error
Pure dipole: Injected current Cohen et al 1990 16 EEG and 16 MEG sensors EEG error: 10 mm MEG error: 8 mm Cuffin 2001, 21-32 electrodes 10.5 mm error with realistic models 10.6 mm with spherical!
Merlet and Gotman 1999 28 scalp EEG (separate recordings) ; during SEEG: 8 scalp EEG for matching epileptic spikes; Spikes modelled by 1 to 3 dipoles Distance between main dipole and main SEEG contact: 11 ± 4 mm Propagation patterns were confirmed Activity from focal sources not visible (at least 8 intracerebral contacts)
Mégevand et al 2013: comparison with irritative and seizure onset zone 38 patients with subdural grids; 64-128 EEG channels Median distance EEG source with grid point : 15 mm No difference in accuracy between patients with temporal or extra-temporal epilepsy Comparison with fmri: Sharon et al 2007, EEG+MEG = 5mm But intrinsically extended sources
b) Detectability
Gavaret et al 2006: frontal epilepsies 64 channel EEG SEEG SEEG Anterior cingulate is well localized Only lateral part of orbito frontal -> architectony seems more important than depth
Gavaret et al 2004: temporal epilepsies 64 channel EEG medial lateral - Patients with purely mesial activity : not visible at surface - Lateral activity can be localized with good precision
b) Network activity
Extracting network is of great interest but gets more difficult Increasing noise Bai and He, 2006: simulation study Difficult to estimate more than 4 or 5 sources active simultaneously.
MEG ICA networks (1) Extraction of epileptic components from surface recordings Malinowska et al Hum Brain Mapp 2013 See also Kobayashi et al 1999, Ossadtchi et al 2004, 2005
MEG MEG ICA networks (2) L SEEG -> very good match between surface and depth recordings Malinowska et al, Hum Brain Mapp 2013
MEG ICA networks (3) MEG L SEEG L Malinowska et al, Hum Brain Mapp 2013 -> surface recording only sees part of the network
Limitations of separate recordings Do not warrant recording the exact same activity Fluctuations can be due to state of vigilance, medication Do not permit using inter-event fluctuations as a source of information on the relations between modalities See EEG-fMRI single-trial evoked potentials studies (Debener et al 2005 etc )
SIMULTANEOUS RECORDINGS
Alarcon et al 1994 Simultaneous scalp EEG foramen ovale Surface recordings are sensitive only to the propagated activity
Lantz et al 2001 22 scalp EEG recorded simultaneously with subdural electrodes Lantz et al 1996, 1997 Epifocus source localization Sources can be localized with sublobar accuracy: medial vs lateral subtemporal; anterior vs posterior lateral temporal
Santiuste 2008 Simultaneous MEG / 1 SEEG electrode Detection of 95 % of neocortical spikes, but only 25 to 60 % of spikes from mesial structures
248MEG +137 SEEG c B GPH PA PI TP Badier et al in prep; Tuesday workshop «The added value of simultaneous multimodal recordings in neurosciences
Simultaneous EEG-MEG-SEEG of Evoked activity Dubarry et al, NIMG 2014 Evoked activity can be captured simultaneously on the three modalities
Single trial analysis Dubarry et al, NIMG 2014 Single-trial activity can be captured thanks to independant component analysis -> this opens the way to investigate trial-to-trial coupling between surface and depth recordings
Application : ICA versus beamformer In a focal case of epilepsy, beamformer and ICA give very similar results for recovering the spikes seen in simultaneously recorded SEEG M. Woodman in prep
CHALLENGING SITUATIONS: DEEP ACTIVITY HIGH FREQUENCIES
Deep Structures: MEG Dalal et al 2013, reading task, zero lag correlation between hippocampus and MEG signals Sylvain Baillet: Comparison of virtual electrode in hippocampus and SEEG recording
Deep structures: EEG Wennberg et al 2011 Spikes in hippocampus; N=43; low amplitude on scalp (5-10 µv) Koessler et al N=368, amplitude 7µV (snr=-2.1db on raw data) See also Merlet et al 1998
High frequency oscillations Rampp et al 2010 simultaneous subdural/meg : high gamma (~80Hz) can be seen on MEG Zelmann et al 2014 simultaneous ECoG/EEG: Ripples fast ripples (100-300Hz) can be seen on surface EEG Recent reports of even high frequencies Xiang et al 2010: 900 Hz Usui et al 2015 >1000 Hz (!)
Biophysical/computational modelling Attal and Schwartz 2013 Biophysical model of hippocampus and amygdala Attal et al 2012: computational modelling for estimating hidden variables
CONCLUSIONS
«Take home» message Intracerebral recordings confirm surface recordings in a large proportion of cases Neocortical sources with sufficient extent (signal to noise increase with averaging) In some favorable cases deep structures can be observed Intracerebral EEG keeps a higher sensitivity (MEG/EEG see the «tip of the iceberg») This can be improved by signal processing and modelling Simultaneous recording and computational/biophysical modelling could improve the understanding of signals and strategies to be used