Source Localization for EEG/MEG Stavroula Kousta Martin Chadwick Methods for Dummies 2007

Source Localization for EEG/MEG

Stavroula KoustaMartin Chadwick

Methods for Dummies 2007

Source Localization: The Question

Is it possible to find the locus of electrical activity in the brain based upon the map of recorded electrical activity?

EEG MEG

Forward Solution

Model Head

Brain

Skull

Scalp

-4 μV

-5 μV

Model Data

0 μV

-1 μV-2 μV

-3 μV

1 μV2 μV

3 μV

4 μV

5 μV

+

Location OrientationAmplitude

Inverse problem

-5 μV

-4 μV

Recorded Data

0 μV

-1 μV-2 μV

-3 μV

1 μV2 μV

3 μV

4 μV

5 μV

Desired Model Solution

Brain

Skull

Scalp

?

?

? ?

?

ILL-DEFINED!!!

+

-

+ -

+-

+ -

Any field potential vector could be consistent with an infinite number of possible dipoles

Specify initial dipole parameters in head model

Generate scalp potentials/fields using forward solution (Vmodel)

Calculate SSE

Σ (Vmodel –Vdata)2 ΔSSE<εAccept dipole parameters

Adjust Parameters

Yes

No

Inverse Solution

Slotnick, 2005

Source Model Initial Parameters

‘Imaging’‘Imaging’‘Equivalent Current Dipoles’ (ECD)

‘Equivalent Current Dipoles’ (ECD)

Head Model

Spherical Head Model

Head Model

3-spheres Head Model*

* Rush & Driscoll, IEEE-TBME, 1969

Head Model

Realistic Geometry Head Model*

* He et al., IEEE-TBME, 1987

Head Model

3-shell BEM RG Head Model*

* Hamalainen & Sarvas, IEEE-TBME, 1989

Head Model

FEM RG Head Model*

* Yan et al. Med. & Biol. Eng. & Compt. 1991

Forward Solution

Source Model (e.g. ECD) + Head Model (e.g. boundary-element):

V = C V + G

Inverse Solution

• Model fitting– Minimize the sum of squares error between

the data and the model by adjusting model parameters (dipole location, orientation, magnitude, time-course).

• Parameter fitting routines– Levenberg_Marquardt method (Marquardt, 1963)– Simplex method (Nelder and Mead, 1965)– Simulated annealing (Kirkpatrick, 1984)– Genetic algorithms (McNay et al., 1996)

EEG & MEG Source Localisation:The Buttons

Components of the source reconstruction process

Source/Head Model

Source/Head Model Forward modelForward model Inverse solutionInverse solutionRegistrationRegistration

1) Source/Head Model

• Select data file

• Select title for the analysis

• Select “MRI”

‘Imaging’‘Imaging’‘Equivalent Current Dipoles’ (ECD)

‘Equivalent Current Dipoles’ (ECD)

• Select the size of your head model

• Select the relevant source model

Head Model

2) Registration

•This stage co-registers the EEG sensory positions with the head model

•Approximately matches anatomical markers between the two spaces

•Followed by a more accurate surface-matching routine.

3) The Forward Model

• Select “Forward Model”

• This will now create a forward model based on the specifications of your source model and your head model.

Important!

Source model

Source model

Forward model

Forward model

Inverse solution

Inverse solution

Registration

Registration

The same for all conditions.

Therefore, only done ONCE for each subject

Repeated for each condition

4) The Inverse Solution

The 3D reconstruction is now complete, and you are ready to perform you statistical analyses.

Sources

• Cuffin (1998) http://ieeexplore.ieee.org/iel4/51/15468/00715495.pdf

• Slotnick, D. (2005). In T.C. Handy (Ed.), Event-related potentials: A methods handbook. MIT Press.

• Mattout, Phillips & Friston (2005) SPM course

http://www.fil.ion.ucl.ac.uk/spm/course/slides05/ppt/MEEG_inv.ppt

• Bahador Bahrami (2006)

http://www.fil.ion.ucl.ac.uk/~mgray/Presentations/Source%20localization%20for%20EEG%20and%20MEG.ppt