Upload
others
View
15
Download
0
Embed Size (px)
Citation preview
EEG and MEG: functional brain imaging with high temporal resolution
Syed Ashrafulla
electrical signals in the brain
Source: Baillet, S., Mosher, J. C., & Leahy, R. M. (2001). Electromagnetic brain mapping. IEEE SIgnal Processing Magazine, (November).
recordable signals Pyramidal Neuron
Soma
Apical Dendrites
Axon
Right Hand Rule
Source:Matti Hamalainen
Murakami, S., & Okada, Y. (2006). Contributions of principal neocortical neurons to magnetoencephalography and electroencephalography signals. The Journal of physiology, 575(Pt 3), 925–36. doi:10.1113/jphysiol.2006.105379
measurable currents
• We can only measure assemblies of neurons.
0.2pAm
10nAm Or 50,000 synchronous cells
Weakest measurable cortical signal Model as one “dipole”
Current dipole of cortical pyramidal cell
Area 0.63mm2
Cortical area
0.9mm
electroencephalography (EEG)
intracellu
lar curren
t extr
acel
lula
r cu
rren
t
extracellular cu
rrent
magnetoencephalography (MEG)
induced magnetic field
MEG signal strength
MEG artifacts
• Use signal space projection and noise cancellation techniques in preprocessing.
what can/can’t be seen
Source: Matti Hamalainen
what can/can’t be seen
Source: Matthew Longo
sensitivity profiles
EE
G
ME
G axial grad
iom
eter M
EG
pla
nar
gra
dio
met
er
ME
G m
agn
eto
met
er
EEG vs MEG
EEG MEG Signal magnitude 10 mV (easy to detect) 10 fT, difficult to detect
Measurement Secondary currents Primary currents
Signal purity Skull/scalp attenuation Little effect of skull/scalp
Temporal Resolution ~ 1 ms ~ 1 ms
Spatial Localization ~ 1 cm < 1 cm
Experimental Flexibility Moves with subject Stationary with subject
Dipole Orientation Tangential & radial Most sources are not fully radial
Only tangential
Source: Matthew Longo
EEG/MEG vs. fMRI
EEG/MEG fMRI Temporal Resolution ~ 1 ms ~ 1 s
Signal Type Direct (currents) Indirect (BOLD)
Signal Reconstruction Ill-posed inversion Deconvolution
Spatial Localization ~ 1 cm ≅ 1 mm for high-T
Sensitivity depth ~ 4 cm Whole-brain
Sensitivity profile drops off as square of distance from sensor
Signal orientation Tangential (and radial) Can cause signal cancellation
Agnostic
resolution comparison
modelling EEG/MEG recordings
1Am
Or in matrix form:
qam1
m2
m3m4
m5
m6
m7
g1
g2
g3g4
g5
g6
g7
m1 = g1 ´1
m2 = g2 ´1
m1 = g1
a ´qa + g1
b ´qb + g1
c ´qc
m2 = g2
a ´qa + g2
b ´qb + g2
c ´qc
qb
qc
m =
m1
m2
é
ë
êêêê
ù
û
úúúú
=
g1
a g1
b
g2
a g2
b
é
ë
êêêê
ù
û
úúúú
1
0
é
ë
êêê
ù
û
úúú
=G
1
0
é
ë
êêê
ù
û
úúú
m =
m1
m2
é
ë
êêêê
ù
û
úúúú
=
g1
a g1
b
g2
a g2
b
é
ë
êêêê
ù
û
úúúú
qa
qb
é
ë
êêêê
ù
û
úúúú
+
n1
n2
é
ë
êêêê
ù
û
úúúú
=Gq+ n
Sensor (measurement) noise
inverse imaging
s1
s2
sn
Assume a density of dipoles oriented normally to the cortical surface. Find their amplitude.
m =Gq+nÞ q̂ = argminq
m-Gq2
q̂ = HmÜ q̂a = Ha
TmÜ Ha = argminh
hTGa=1
hT Cov m( )( )h
Minimum-norm estimation (MNE):
Minimum-variance beamforming (LCMV):
(Hui2010, NeuroImage)
(Hauk2004, NeuroImage)
time-frequency decompositions
Ä
X st
=
t
f
Cstf = X st *wtf
θ band: 4-7Hz α band: 8-14Hz β band: 15-30Hz γ band: 30-100Hz
s: spatial index t: temporal index f: frequency index Source: Dimitrios Pantazis
connectivity & EEG/MEG • Connectivity = Relation between x[n] and y[n] Correlation: do x[n] and y[n] change at the same time?
Coherence: do x[n] and y[n] change at the same time for frequency f?
Causality: Is y[n] required to generate x[n]?
Granger causality: Does y[n-1] help predict x[n]?
All methods benefit from the high temporal resolution of EEG/MEG.
BrainStorm demonstration