Module 25:
Effective Connectivity
Brain Connectivity
• Effective Connectivity
– Directed influence of one brain region on the
physiological activity recorded in other brain regions.
– Claims to make statements about causal effects among
tasks and regions.
– Usually makes anatomically motivated assumptions and
restricts inference to networks comprising of a number of
pre-selected regions of interest.
V5
V1 PPC
Effective Connectivity
• Methods include:
– Structural Equation Modeling
– Granger Causality
– Dynamic Causal Modeling
– Bayes Net
SEM
• Structural Equation Models comprise a set of
regions and a set of directed connections.
bAC
A" C"
bAB bBC
B"
• Path coefficients defined between pairs of nodes.
• Directional relationships are assumed a priori.
– Often given a causal interpretation.
Example
bAC
A" C"
bAB bBC
B"
! y $ ! $! y $ ! e(1) $
# A & # 0 0 0 &# A & # &
# yB & = # bAB 0 0 &# yB & + # e(2) &
# & # &# & # e(3) &
#" yC &% " bAC bBC 0 %#" yC &% " %
yt = Myt + et et ~ N (0, R) t = 1,…T
Set-Up
• We can rewrite:
yt = Myt + et
as
yt = ( I − M ) −1 et
• Hence, we can write the covariance matrix of yt as
Σ(θ ) = ( I − M ) −1 R((I − M ) −1 )T
• The parameters θ are the unknown elements of
the matrices M and R.
Estimation"
• The covariance of the data represents how
activities in two or more regions are related.
• In SEM we seek to minimize the difference
between the observed covariance matrix and
the one implied by the structure of the model.
– The parameters of the model are adjusted to
minimize this difference.
– Typically maximum likelihood estimation is used to
estimate the parameters.
Dynamic Casual Modeling
• DCM attempts to model latent neuronal
interactions using hemodynamic time series.
– Based on a neuronal model of interacting regions,
supplemented with a forward model of how neuronal
activity is transformed into the observed response.
• Effective connectivity is parameterized in terms of
the coupling among unobserved neuronal activity
in different regions.
– We can estimate these parameters by perturbing the
system and measuring the response.
Illustration
y"1" y"2"
Hemodynamic
Neuronal
u1" Z"1" Z"2"
u2"
Neuronal Model
• Define the neuronal states as:
T
z = (z1 ,… z N )
• The effective connectivity model is described by:
" J %
zt = $$ A + ∑ ut ( j)B j '' zt + Cut
# j=1 &
where zt is the neuronal activity at time t (latent)
and ut(j) is the jth of J inputs at time t (known).
Interpretation
• The matrix A represents the first order connectivity
among regions in the absence of input.
– Specifies how regions are connected and whether these
connections are uni- or bidirectional.
• The matrix C represents the extrinsic influence of
inputs on neuronal activity.
– Specifies how inputs are connected to regions.
• The matrices Bj represent the change in coupling
induced by the jth input.
– Specifies how connections are changed by inputs.
" J %
zt = $$ A + ∑ ut ( j)B j '' zt + Cut
# j=1 &
y"1" y"2"
Hemodynamic
Neuronal
u1" Z"1" Z"2"
z1 = a11z1 + a12 z2 + c11u1
u2"
z2 = a21z1 + a22 z2 + b212u2 z1
Hemodynamic Model
• Neuronal activity causes changes in blood
volume and deoxyhmoglobin that cause
changes in the observed BOLD response.
• The hemodynamics are described using an
extended Balloon model, which involves a set
of hemodynamic state variables, state
equations and hemodynamic parameters θh.
Extended Balloon Model
Activity-dependent signal: s! = z − κs − γ ( f − 1)
Flow induction: f! = s
Changes in volume: τν! = f −ν 1/ α
Changes in dHb: τq! = fE( f , ρ ) ρ −ν 1/ α q ν
Hemodynamic response y = λ (ν , q)
State Equations
Neuronal state:
Neuronal activity - zt with parameters θc.
Hemodynamic states:
Vasodilatory signal - st
Inflow - ft
Blood volume - vt
Deoxygenation content - qt
The observed data: yt=λ(qt,vt) with parameters θh.
Bayesian Analysis
• Combining the neuronal and hemodynamic states
x={z, s, f, v, q} gives us the following state-space
model:
x! = f ( x, u,θ )
y = λ ( x, θ )
• Analysis performed using Bayesian methods
– Normal priors are placed on θ.
– The posterior density is used to make inferences about
the connections.
End of Module
@fMRIstats