Anatomically and Functionally Constrained MEG/EEG Source Estimates Matti Hämäläinen MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging Charlestown, MA USA
Contents • Introduction: Multimodal imaging, MEG and EEG • Anatomically and functionally constrained source estimates • Recent developments Matti Hämäläinen 7/2009 2
Noninvasive Multimodal Brain Imaging Estimated neural currents EEG MEG fMRI NIRS MRI Anatomy 3 Matti Hämäläinen 7/2009
MEG and EEG
MEG and EEG EEG EEG • The primary current is related MEG V V to the postsynaptic activity • The primary current generates a potential distribution (EEG) and the associated volume currents J p J p J p • The primary and volume currents together also create a magnetic field (MEG) • However, the net effect of volume currents is rather straightforward to take into account in MEG whereas the it is difficult to compute the EEG potential distribution accurately Matti Hämäläinen 7/2009
Realistically-shaped forward models for MEG and EEG MEG ≈ EEG ≠ Multilayer model: Homogeneous model: skull and scalp taken into account, skull taken as an insulator, conductivities needed result independent of conductivity Theoretical analysis: Hämäläinen and Sarvas, 1989 Experimental validation: Okada et al. , 1999 Matti Hämäläinen 7/2009 6
Primary currents in the cortex cortex MEG = 0 EEG � 0 MEG � 0 EEG � 0 current sources B = 0 Primary currents No magnetic field from radial currents in the sphere model Matti Hämäläinen 7/2009 7
Tangential, radial, and tilted sources MEG tangential radial tilted EEG tangential radial tilted MEG has only one prototypical field pattern Matti Hämäläinen 7/2009 8
MEG and EEG sensitivity to cortical sources MEG EEG Matti Hämäläinen 7/2009 9
Anatomically and functionally constrained source estimates
Motivation to use distributed source models • Account for non-focal sources • Automatic analysis without heuristic choices often needed in multidipole models • Incorporate anatomical and functional MRI constraints • Lower SNR data can be processed Transformation of data to brain space without strong assumptions about the sources Matti Hämäläinen 7/2009 11
Minimum-Norm Solutions Data and noise: y = Gq + � E { �� T } = C Fit the data with a source penalty term: || y − Gq || 2 � � C + || q || p ˆ q = argmin q R – Minimum-norm estimates (MNE): p = 2 – Minimum-current estimates (MCE): p = 1 Matti Hämäläinen 7/2009 12
������������������������������������������ �������� Retinotopic mapping with MNE Ahlfors et al. 1992 Matti Hämäläinen 7/2009 13
Modern MNE • Source locations (and orientations) constrained to the cortical mantle • Forward solution with BEM • Full noise-covariance matrix estimates computed from raw data • Display on an inflated cortex to reveal the sulci • Compute statistics • Combined MEG and EEG solutions • fMRI-guided solutions Dale et al. 2000 Matti Hämäläinen 7/2009 14
Cortical Source Location Constraints Tessellation of the cortex: For source estimation, the surface Source location and orientation is typically decimated, resulting information in 6000 - 10000 source locations Matti Hämäläinen 7/2009 15
Inflated Cortex Topologically correct tessellation can be inflated Dale, Fischl, Sereno et al. Matti Hämäläinen 7/2009 16
Inflation to a Sphere and Registration Individual Aligned with average brain Align sulcal patterns to the average brain MEG activity estimate Mapped to the average brain Morph Matti Hämäläinen 7/2009 17
Noise normalization • Convert the current values into a test statistic – dSPM (Dale et al. ) – sLORETA (Pascual-Marqui et al. ) • Divide the current with its standard deviation • Analyze MEG/EEG data like fMRI or PET Dale et al. 2000 Matti Hämäläinen 7/2009 18
MNE and dSPM MNE dSPM - Auditory MEG data - Source locations constrained to the cortex - No orientation constraint - dSPM and sLORETA produce very similar results with real data Matti Hämäläinen 7/2009 19
Loose orientation constraint • Penalize current components tangential to the cortex • Takes the finite spacing between elementary sources into account cortex MEG = 0 EEG � 0 MEG � 0 EEG � 0 current sources Lin et al. 2006 Matti Hämäläinen 7/2009 20
Effect of additional constraints Location of simulated source MNE dSPM Unweighted Depth-weighted Depth-weighted and LOC Matti Hämäläinen 5/2009 21
Spatial dispersion of cortically-constrained MEG and EEG source estimates MEG EEG MEG+EEG 0 cm 2 cm 4 cm Molins et al. 2008 Matti Hämäläinen 7/2009
Comparison of MEG, EEG, and fMRI (dSPM) fMRI MEG EEG MEG + EEG Sharon et al. 2007 Matti Hämäläinen 7/2009 23
Visual percepts of an ambiguous scene MEG signals at an occipital sensor 12 Hz 30 Spectral density Hz) 15 Hz � 20 (fT/cm/ 10 30 0 10 20 Time / s -1 0 1 2 -2 Noise: 12 Hz 15 Hz 20 Hz 15 Hz 12 Hz 10 Hz Percept ‘vase’ Percept ‘faces’ Matti Hämäläinen 7/2009 Parkkonen et al., PNAS , 2008 24
Extract tag-related activity: MNE + GLM Behavioral report t > 2 s ’vase’ ’faces’ 0.25 s T T Source waveform = = GLM T T a V a F 12.0 Hz tags b V b F 15.0 Hz alpha 9.5 Hz 10.0 Hz mu * * 18.9 Hz mains 50.0 Hz linear trend 1 s Matti Hämäläinen 7/2009 25 Parkkonen et al., PNAS , 2008
Group analysis Significant activity in either tag frequency Amplitude ratio at ROI 2.0 vase-tag / face-tag Amplitude ratio 100% 1.5 75% 50% N = 8 1.0 'faces' 'vase' Percept Left Right Matti Hämäläinen 7/2009 Parkkonen et al., PNAS , 2008 26
fMRI-guided estimates • Prioritize locations of significant fMRI activity (increase source variance) • fMRI incorporated as a constraint, not an integrated analysis procedure Dale et al. 2000 Matti Hämäläinen 7/2009 27
Going further: the fIRE Model fMRI-Informed Regional Estimates Time: • Region-specific, independent neural and vascular time cources Space: • Time courses modulated by a scalar z • z is location specific and smooth in space Ou et al. 2009 Matti Hämäläinen 7/2009 28
FIRE: Graphical Model Brain activity, space Current sources, space & time fMRI data, space & time EEG/MEG data, space & time Neural waveform, time Vascular waveform, time Ou et al. 2009 Matti Hämäläinen 7/2009 29
Scenario One: No Silent Sources Simulation MNE fMNE fARD FIRE Simu. MNE fARD FIRE fMNE 25 Matti Hämäläinen 7/2009
Scenario Two: Silent Vascular Activity Simulation MNE fMNE fARD FIRE Simu. MNE fARD FIRE fMNE 26 Matti Hämäläinen 7/2009
Scenario Three: Silent Neural Activity Simulation MNE fMNE fARD FIRE Simu. MNE fARD FIRE fMNE 27 Matti Hämäläinen 7/2009
FIRE: Features and Benefits • Avoids excessive bias towards fMRI – If the hemodynamic signal is missing, defaults to the L 2 minimum-norm solution – If the neural signal is missing does not attempt to imply neural signals at the fMRI-only regions • Regional approach: computationally tractable Matti Hämäläinen 7/2009 33
Multimodal imaging involving MEG and EEG • MEG and EEG have complementary strengths • Anatomical MRI – Visualization in anatomical context – Modeling constraints for improved accuracy and to make the source estimation problem less ill posed – Cortical surface visualization common with fMRI: easy comparison, common group analysis approaches • Functional MRI – Comparison of data in the same format – fMRI-informed MEG/EEG estimates to enhance spatial resolution – Goal: Integrated analysis Matti Hämäläinen 7/2009 34
Seppo Ahlfors www.nmr.mgh.harvard.edu Jack Belliveau Anders Dale (UCSD) Bruce Fischl Polina Golland (MIT/CSAIL) Riitta Hari (HUT) Fa-Hsuan Lin Maria Mody Antonio Molins Wanmei Ou (MIT/CSAIL) Lauri Parkkonen (HUT) Tommi Raij Bruce Rosen Dahlia Sharon (Stanford) Daniel Wehner Thank you! Thomas Witzel Matti Hämäläinen 7/2009 35
Recommend
More recommend
