You are on page 1of 50

Connectivity Modeling

Anthony Randal McIntosh

Department of Psychology
University of Toronto
Overview
• Theoretical issues and basis for analytic approach
• Structural equation modeling
– Integrating anatomy and function
• Partial Least Squares
– Identification of distributed systems
• Applications
– Sensory learning
– Working memory
Theoretical focus
• All behavioural and cognitive operations in the
brain come about through the the action of
distributed networks
• In order to assess this, need methods that can
measure the whole brain and analyses that look at
more than one region at a time
• Ideally, we would like to analyze spatial and
temporal patterns of brain function at the same
time
Causal patterns in brain research

Design/Task Brain Behaviour


Causal patterns in brain research
Response

Stimulus
Causal patterns in brain research
Response

Stimulus
Theory to Analysis
• Examine the influences between brain areas
– Interregional correlation (Horwitz, et al, 1984)
– Structural equation modeling (McIntosh & Gonzalez-Lima, 1991,
Buchel & Friston, 1997)
– Multiple regression and extensions (e.g., Kalman filters, Buchel &
Friston, 1998)
– Bayes networks (Dynamic Causal Modeling, Friston, Penny, et al,
2003)
• Identification of interacting regions
– Partial Least Squares (McIntosh, Bookstein, et al, 1996)
– Canonical Variates Analysis (Strother et al, 1995)
– Independent Components Analysis - 32 flavours (McKeown et al,
1998, Calhoun et al, 2001, Beckmann, Smith, et al., 2002)
Functional and Effective Connectivity
Structural Equation Modeling
• Multivariate multiple regression
• Combines interregional covariances with
anatomical framework
• Provides means to assess whether effective
connections are modified by task-demands or
differ between groups
• Is not meant to be a model test in the coventional
sense
– Goodness of fit not as relevant
Structural Equation Modeling
w = 0 .6 1 1 A B C D
A x = 0 .0 1 1 B
A 1 .0 0
y = 0 .6 1 4 z = -0 .5 5 3
B 0 .4 8 1 .0 0

C 0 .6 2 0 .1 6 1 .0 0

C D D 0 .2 4 -.4 1 0 .0 6 1 .0 0

S tr u c tu r a l E q u a tio n s

A = xB + yC +  

B = w A + zD +  B
Dorsal vs. Ventral Cortical Visual
Streams
O b je c t Id e n tific a tio n S p a tia l L o c a tio n

7
19d

4 6 /4 7 4 6 /4 7
17 19v 17 19v
18 18
37 37 21
21

P a th C o e ffic ie n ts
P o s itiv e N e g a tiv e
0 .7 to 1 .0
0 .4 to 0 .6
0 .1 to 0 .3
0

McIntosh et al, J. Neurosci, 1994


What inferences does Structual
Equation Modeling allow?
Object Space

46 46 46
46

21 21 21 21

7 7 7 7
37 37 37 37
19d 19d 19d 19d
19v 19v 19v 19v

P o s itiv e N e g a tiv e
0 .6 5 - 1 .0
0 .3 5 - 0 .6 5
0 .1 - 0 .3 5
0
What inferences does Structual
Equation Modeling allow?
Object Space

46 46 46
46

21 21 21
21

7 7 7 7
37 37 37
37
19d 19d 19d 19d
19v 19v 19v
19v

P o s itiv e N e g a tiv e
0 .6 5 - 1 .0
0 .3 5 - 0 .6 5
0 .1 - 0 .3 5
0
Partial Least Squares
• “Least-squares” decomposition of “part” of a covariance
matrix
• PLS is optimized to explain the relation between two or
more blocks of data
– what pattern in one block most strongly covaries with a pattern in
another block?
• Ignores the relation among items within data blocks
• Statistical assessment through resampling algorithms
– Permutation test and bootstrap estimation of standard error

McIntosh, Bookstein, et al, Neuroimage, 1996


Task PLS
Compute matrix Mdev which is n*k by m
n is the number of subjects or repetitions,
k is the number of scans,
m is the number of voxels.
Each voxel is centered relative to the grand mean.
• Compute matrix X, a matrix of scan means expressed as deviations
from the grand mean.
Alternative: project (correlate) set of orthonormal contrasts (design
matrix) on to the data matrix
Matrix X now the covariance of image activity with the experimental
design.
Task PLS
• Perform a singular value decomposition (SVD) on X to define the
latent variables (LV):
SVD(X) = [U,S,V] where:
X = U*S*VT
U is the k by m orthonormal matrix containing voxel weights
(singular or eigen image).
S is a diagonal matrix of k singular (“eigen”) values. (The kth
singular value is zero because we use deviation values in X,
i.e., we eliminate the grand mean to create X).
VT is the transpose of matrix V, a k by k orthonormal matrix of
scan weights.
• Project singular image on to original data to obtain “brain scores”
Index of how well each subject shows the effect
Task PLS
Behavior PLS
• Compute matrix M which is n*k by m
n is the number of subjects/repetitions,
k is the number of scans,
m is the number of voxels.
• Create vector B, which is an n* k by 1 vector of
performance measures for each scan.
• Create matrix Y, which contains the scan-specific
correlations of voxel activity (Yk) and behavior
(Bk).
Behavior PLS
• Perform a singular value decomposition (SVD) on
Y to define the latent variables (LV):
SVD(X) = [U,S,V] where:
X = U*S*VT
U is the k by m orthonormal matrix containing
voxel weights (singular or eigen image).
S is a diagonal matrix of k singular (“eigen”)
values.
VT is the transpose of matrix V, a k by k
orthonormal matrix of scan weights.
Behavior PLS
C r o s s - c o r r e la t io n o f B a n d M ,
M a trix Y

B r a in Im a g e s
B e h a v io r
M a tr ix M
M e a s u re s
M a t r ix B

C r o s s -c o r r e la tio n o f B a n d M ,
S1
M a trix Y

V1 U 1
S in g u la r V a lu e D e c o m p o s it io n (S in g u la r Im a g e )
o f M a trix M

Correlation of Brain
0

Scores & Behavior


-1
Brain Scores
-2

-0 . 6 -0 .4 -0 . 2 0 .0 0 .2 0 .4 0 .6
4

-2

-4

-1 .0 - 0 .5 0 .0 0 .5 1 .0 1 .5
8

U1
6

(S in g u la r Im a g e ) 2

B r a in Im a g e s -2

-4

M a tr ix M - 1 .0 -0 .5 0 .0 0 .5 1 .0 1 .5 2 .0

B e h a v io r M e a s u r e
Statistical Assessment
• Assessment of omnibus/latent variable structure
through permutation tests
– Is the latent variable significantly different from “noise”?
• Assessment of the precision of estimates derived from
PLS through bootstrap estimation of standard errors
– How reliable is the answer?
• Procrustes rotation to original solution space used to
correct for axis rotation and reflection during
resampling
Milan & Whittaker, 1995, Royal Stat Society J
Why use bootstrap?
0 .0 2

0 .0 1 5
Permutation
• Estimation of standard errors is a direct
0 .0 1
Est Standard Error

0 .0 0 5

assessment of the stability of your data


0
0 .0 2

0 .0 1 5
Bootstrap
– A signal can be significantly different from
noise (e.g., P<0.01), but not be reliable
0 .0 1

0 .0 0 5
0
0 .0 2

0 .0 1 5
ANOVA
0 .0 1

0 .0 0 5
0
-0 .0 4 0 0 .0 4

Singular Vector Weight


Causal patterns in brain research
Response

Stimulus
Multiblock PLS
A v e r a g e d e v ia tio n w ith in s c a n ,
M a tr ix X
N o r m a liz e r o w s to u n it-le n g t h

B r a in Im a g e s G ra n d
M a tr ix M M ean

C r o s s - c o r r e la t io n o f B a n d M ,
M a trix Y
N o r m a liz e r o w s to u n it-le n g t h

B e h a v io r B r a in Im a g e s
M e a s u re s M a tr ix M
M a tr ix B
Multiblock PLS
A v e r a g e d e v ia t io n w it h in s c a n ,
M a trix X

S1
C r o s s - c o r r e la tio n o f B a n d M ,
M a trix Y
V1 U 1
( S in g u la r Im a g e )

M a t r ix Z
S in g u la r V a lu e D e c o m p o s it io n
o f M a tr ix Z

and behavior
Within-task correlation of
Average Brain Score

within-task

brain scores
U1
( S in g u la r Im a g e )

B r a in Im a g e s
M a tr ix M
How do we use this?
Ta rg e t D is tr a c to r

U n p a ir e d tr ia ls
P a ir e d t r ia ls
560
T o n e = 1 k H z F M ~ 6 5 d B - 5 0 0 m s d u r a tio n

R e a c tio n T im e (m s e c )
540
1 D is tr a c to r
520

Tone P ( V is u a l/T o n e 2 )= 0 .2
500
2 lo w P
480

460
To n e
Scans 3 h ig h P 440

420

Tone To n e To n e To n e Ton e D is tr a c to r
4 h ig h P lo w P h ig h P h ig h P h ig h P
2 3 4 5 6
Scan
To n e
5 h ig h P
P ( V is u a l/T o n e 1 ) = 0 .7
6 D is tr a c to r

McIntosh, Cabeza & Lobaugh, J Neurophys 1998


Sensory Associative Learning
1 Identify system(s) that respond to change
in significance of the tone
2 Identify system(s) that relate to (effect) a
change in behavior as a result of learning
3 Identify the overlap between 1 and 2
Task PLS
-2 8

-4

+20

16

14
B ra in s c o re s

12

10

2
VD1 TLP THP1 THP2 THP3 VD2
Scan
Behaviour PLS
-2 8

-4

+20
B r a in S c o r e s

TLP THP2

VD2

THP1 THP3

B e h a v io r
Multiblock PLS

TLP THP2
2
B r a in S c o r e s

VD2
B r a in S c o r e s

-1
THP1 THP3

-2

-3
TLP THP1 THP2 THP3 VD2
B e h a v io r S c o r e s
Explaining regional activation

Task B e h a v io u r
A d ju s te d rC B F

A d ju s te d r C B F
To n e - sc a n 1 To n e - sc a n 2 To n e - s ca n 3

V is u a l Ton e Tone To n e Ton e V is u a l


Scan 1 S can 2 S can 3 Scan 4
Scan To ne sca n 4 V is u a l
R T - D iffe r e n c e (u n p a ir e d - p a ire d )
Seed voxel PLS
-4

+20

TLP r = -.5 7 THP1 r = -.1 5


A d ju s te d r C B F

T H P 2 r = -0 .2 2 T H P 3 r = 0 .7 2

B ra in S c o r e s
Structural Equation Model P o s itiv e
10 0 .1 - 0 .3 5

6 0 .3 5 - 0 .6 5
0 .6 5 - 1 .0

N e g a tiv e

42 0 .1 - 0 .3 5
0 .3 5 - 0 .6 5
0 .6 5 - 1 .0

18
18 TLP 0

10 10 10
6 6 6

42 42 42

18 18 18
18 THP1 18 THP2 18 THP3
ERP/MEG/fMRI Data Sets
Occasions, Trials,
Subjects, Groups

I M E
T
SPACE
Voxels, detectors, electrodes
ERP/MEG/fMRI Data
Occasions,Trials,
Subjects, Groups Flatten the matrix

me
T i
Space: Voxels/Channels
Occasions,Trials,
Subjects, Groups

Time and Space


Task PLS
Spatiotemporal PLS- fMRI

Voxel B r a in Im a g e s
Saliences M a tr ix Y Temporal Brain Scores

How strongly does


the brain differentiate
tasks at each
timepoint???
McIntosh, Protzner & Chau, Neuroimage, in press
N-back Task Variant
Motivation
• Working memory may be conceived as the
interplay of sustained attention and memory
• For a given WM task - is there a
dissociation that would reflect the relative
contribution of an attentional component
v.s. recruitment of memory retrieval?

Lenartowicz & McIntosh, submitted


N-Back Task Variant
2-back (standard - Std)

0-back (detection)

1-back

2-back (Cued)
Time
Task PLS
0

2 8
6

B o o ts t r a p R a tio
4
4
T im e ( s e c )

2
6

B r a in S c o r e
0 0

8 -2

-4

10 -4
-6

-8
12 0 2 4 6 8 10 12 14
-8 T im e ( s e c )
0 -b a c k
14 1 -b a c k
+16 +24 +32 +40 +48 +56 +64 C u e d 2 -b a c k
S td 2 -b a c k
Task PLS
0

10
2
8

4 6

B o o ts tr a p R a tio
5
T im e ( s e c )

B r a in S c o r e
6
2
0
8 0

- 2-

10 -5
0 2 4 6 8 10 12 14

12 T im e ( s e c )

-1 0
14 0 -b a c k
1 -b a c k
-1 2 0 +12 +24 +36 +48 +60 C u e d 2 -b a c k
S td 2 -b a c k
Task PLS

0 .3

0 .2

0 .1

- 0 .1

- 0 .2

- 0 .3

- 0 .4
0 2 4 6 8 10 12 14

0 -b a c k
1 -b a c k
C u e d 2 -b a c k
S td 2 -b a c k
Task PLS

0 .2 5

0 .2

0 .1 5

0 .1

0 .0 5

-0 .0 5

-0 .1

-0 .1 5
0 2 4 6 8 10 12 14

0 -b a c k
1 -b a c k
C u e d 2 -b a c k
S td 2 -b a c k
Task PLS
• Dominant effect on anterior cingulate
activity
• What is the functional connectivity of the
anterior cingulate?
– Does it change between the 2-back tasks?
• Is the pattern of functional connectivity
related to behavior?
Behaviour/Seed PLS 15

10 1

0 .8

0 .6

C o r r e la tio n
0 .4

5 0 .2

-0 .2

-0 .4
0
-0 .6

-0 .8

-1
C u e d 2 -b a c k S td 2 -b a c k
-5
AC
RT
-1 0 H its
-1 6 -4 +8 +20 +32 +44 +56 +68
Behaviour/Seed PLS
10
1

0 .8

0 .6

C o r r e la t io n
5 0 .4

0 .2

0 -0 .2

-0 .4

-0 .6

-0 .8

-5 -1
C u e d 2 -b a c k S td 2 -b a c k

-1 0 AC
RT
-1 6 -4 +8 +20 +32 +44 +56 +68 H its
Summary & Implications
• Anterior cingulate activity differentiates tasks based
on attentional demands
• Functional connectivity varies with attentional
demand
• Relation of functional connectivity patterns to
performance also varies with attentional demand
• Behavioural relevance of a region to a cognitive
operation depends on its pattern of functional
connectivity
– Neural Context - McIntosh, Neural Networks 2001
Evaluating the analytic tools:
How do we know when the math is right?
• Neurobiological interpretation
– Identification of new principles
• Psychological interpretation
– What is the level of nervous system operation that best
relates to the cognitive operation?
• What is the question?
– Level of the answer
– Does the experimental design require a certain analytic
tool?
– Is causality necessary or will correlation suffice?
Thank You

Acknowledgements:
Collaborators: NJ Lobaugh, MN
Rajah, CL Grady
Funding: CIHR, NSERC, JS
McDonnell Fnd

http://www.rotman-baycrest.on.ca
What inferences does Structual
Equation Modeling allow?

You might also like