Transcript Document

We show how a parameterization of
the strongly folded boundary
between the gray and white matter
can be used as constraints for a
beamformer in order to estimate
local activations inside the brain
during a certain task on a time scale
of milliseconds.
Each pair (ψ,φ) on the sphere corresponds to a location in 3-d space
described by its cartesian coordinates (x,y,z). Color codes below are
the values of these coordinates as functions of the angles in
rectangular and polar plots. Contour lines at constant values of x, y
and z define the boundary lines between the gray and white matter
in corresponding sagittal, coronal and axial slices, respectively.
ψ
x-coordinate
z-coordinate
φ
-6.6cm
0
6.6cm
-7.6cm
0
7.6cm -9.8cm
0
9.8cm
Applying the beamformer allows us to estimate the
source power at locations posterior and anterior of the
central sulcus. Going top to bottom in the plots below
means moving left to right along the sulcus. Each row
exhibits the forward solution, the beamformer pattern
and two time series. Dotted blues lines show the time
course from a single MEG sensor as a reference. The
red curves represent the time dependence of local
activity for a time span of 480ms with the vertical
black line at maximum finger flexion. From bottom
to top activation shifts from a time point prior to peak
flexion to a time thereafter, representing traveling
waves from right to left along both walls of the central
sulcus.
φ
y
An MRI scan is transformed
into a coordinate system
defined by the nasion, and the
left and right preauricular
points on the subject’s head.
y-coordinate
The sensitivity of EEG and MEG to sources on the gray-white
matter boundary is quite different. For MEG the orientation is
most important and regions of high sensitivity (indicated in yellow
for MEG and EEG) are located in the walls of the fissures where
the current flow is tangential with respect to the surface defined
by the sensors. For EEG distance from the electrodes is more
important than orientation. The plot on the right shows the
differences of normalized sensitivities with red/yellow indicating
regions where MEG is more sensitive, plotted in blue shades are
areas of higher sensitivity for EEG.
The gray-white matter boundary defines the locations and
directions of the primary currents in the cortex and can be used as
an anatomical constraint for various kinds of beamforming
algorithms [4]. Here we show results using an extension of the
algorithm known as SAM (synthetic aperture magnetometry) [5]
which can be applied to averaged data. A beamforming filter HΘ is
calculated by minimizing the power from all locations and
directions while keeping the signal constant from a location and
direction of interest Θ =Θ(x,y,z,ψ,φ). In general the beamformer
and the power at Θ are given by
ψ
x
Freesurfer [1,2] is used to extract the surface which
defines the gray-white matter boundary and to inflate it
into a sphere.
Sagittal
Coronal
Axial
That way, we have a quasi-continuous representation of the brain
surface and can sample and tessellate it at any desired accuracy.
The transformation into a sphere is unique and
invertible, i.e. every point on the gray-white matter
boundary corresponds to a single point on the sphere
and vice versa because both surfaces are singly
connected and therefore topologically equivalent [3]. The
spherical coordinate system of latitude ψ and longitude
φ can now be mapped onto the brain surface.
The vectors perpendicular
to this surface are of
particular interest because
due to the columnar
organization of the cortex,
the primary currents are
oriented in these directions.
1
C 1Gθ
2
1


Hθ 
and
S

G

C
G
θ
θ
θ
1


Gθ  C Gθ
where C represents the covariance matrix of the data and GΘ is
the forward solution from Θ. For averaged data C cannot be
inverted and the procedure has to be restricted to the signal
subspace by expanding GΘ and HΘ into the eigenvectors υ(k) of C.
The expansion coefficients hk for the beamformer and the source
power can then be expressed in terms of gl and the eigenvalues λl
of C.

gk 
hk =
 
λk  l=1 λl 
M
2
gl
-1
and
2
Sθ
-1


=   .
 l=1 λl 
M
2
gl
References
[1] Dale A.M., Fischl B.,Sereno M.I., Neuroimage 9: 179-194 (1999)
[2] Fischl B., Sereno M.I., Dale A.M,, Neuroimage 9: 195-207 (1999)
[3] Fischl B., Sereno M.I., Tootell R.B.M., Human Brain Mapping 8: 272-285 (1999)
Comparison of fMRI and MEG activity recorded during an experiment
where subjects were asked to synchronize with an external metronome.
Left: Signal intensity masked with a correlation threshold of 0.5.
Middle:Signal intensity at locations along the gray-white matter boundary.
Right: Global power calculated from the beamformer using MEG data
recorded from the same subject performing the same task.
[4] Dale A.M., Sereno M.I., Journal of Cognitive Neuroscience 5/2: 162-176 (1993)
[5] Robinson S.E., Vrba J., in: Recent Advances in Biomagnetism, Tohoku University
Press, Sendai (1999)
Acknowledgement
Work supported by NINDS (grant NS39845), NIMH (grant MH42900) and the Human
Frontier Science Program.