Dynamic Causal Modeling (DCM) A Practical Perspective

1
Dynamic Causal Modeling (DCM) A Practical
Perspective

• Ollie Hulme
• Barrie Roulston
• Zeki lab

2
Disclaimer
The following speakers have never used DCM. Any
impression of expertise or experience is entirely
accidental.
3
Structure

• 1. Quick recap on what DCM can do for you.
• 2. What to think about when designing a DCM
  experiment
• 3. How to do DCM. What buttons to press etc.

4
A Re-cap for Dummies You can ask different
types of questions about brain processing. Questi
ons of Where Questions of How
5
Functional Specialization is a question of Where?

• Where in the brain is a certain
  cognitive/perceptual attribute processed?
• What are the Regionally specific effects

• ?your normal SPM analysis (GLM)

6
Functional Integration is a question of HOW
How does the system work?
What are the inter-regional effects? How do the
components of the system interact with each other?
Experimentally designed input
7
2 Categories of Functional integration analysis
Functional connectivity the temporal
correlation between spatially remote areas
Effective connectivity the influence one area
exerts over another
MODEL-FREE
MODEL-DEPENDENT Hypothesis driven
PPI
DCM!

8

• DCM overview
• DCM allows you model brain activity at the
  neuronal level (which is not directly accessible
  in fMRI) taking into account the anatomical
  architecture of the system and the interactions
  within that architecture under different
  conditions of stimulus input and context.
• The modelled neuronal dynamics (z) are
  transformed into area-specific BOLD signals (y)
  by a hemodynamic forward model (?).

The aim of DCM is to estimate parameters at the
neuronal level so that the modelled BOLD signals
are most similar to the experimentally measured
BOLD signals.
9
Planning a DCM-compatible study

• Experimental design
• preferably multi-factorial (e.g. at least 2 x 2)

1.Sensory input factor At least one factor that
varies the sensory input changing the stimulus
a perturbation to the system
2. Contextual factor At least one factor that
varies the context in which the perturbation
occurs. Often attentional factor, or change in
cognitive set etc.
10
Planning a DCM-compatible study

• TR should be as short as possible lt 2 seconds

2

• Timing problems in DCM
• Due to the sequential acquisition of multiple
  slices there will be temporal shifts between
  regional time series which lie in different
  slices. This causes timing misspecification. At
  short TRs this is not too much of a problem
  since the information in the response variable is
  predominantly contained in the relative
  amplitudes and shapes of hemodynamic response
  rather than their timings. Consequently DCM is
  robust against timing errors up to 1 second

slice acquisition
1
visualinput

• Possible corrections for longer TRs
• 1. slice-timing
• 2. Restrict model to proximate regions. The
  closer they are along z axis the lower the
  temporal discrepancy

11

• Hypothesis and model
• define specific a priori hypotheses.
• DCM is not exploratory!

Specify your hypotheses as precisely as possible.
This requires neurobiological expertise (the fun
part) read lots of papers! Look for convergent
evidence from multiple methodologies and
disciplines. Anatomy is your friend.
12
Defining your hypothesis
Hypothesis A attention modulates V5 directly
When attending to motion.

Parietal areas
V5

Hypothesis B Attention modulates effective
connectivity between PPC to V5
V1
13

• Which parameters do you think are most relevant?
• Which parameters represent my hypothesis?
• Which are the most relevant intrinsic anatomical
• Connections?
• Which are the most relevant changes in effective
• connectivity/connection strength ?
• Which are the relevant sensory inputs ?
• 2. Defining criteria for inference
• single-subject analysis
• What statistical threshold? What contrasts?
• group analysis Which 2nd-level model?
• Paired t-test for parameter agt parameter b,
• One-sample t-test parameter a gt 0
• rmANOVA (in case of multiple sessions per
  subject)
• 3.Ensure that the model you generate is able to
  test your
• hypotheses

14
4.Evaluate whether DCM can answer your
question Can DCM distinguish between your
hypotheses?
Parietal areas
V5
Direct influence
V1
DCM cannot distinguish between direct and
indirect! Hypotheses of this nature cannot be
tested
In case of
15
1.Specify your main hypothesis and its competing
hypotheses as precisely as possible using
convergent evidence from the empirical and
theoretical literature 2.Think specifically about
how your experiment will test the hypothesis and
whether the hypothesis is suitable for DCM to
test. 3.Klaas emphasises that you should Test
your model before conducting the experiment
using synthetic data. Simulation is the key! 4.
DCM is tricky, ask the experts during the design
stage. They are very helpful.
16
A DCM in 5 easy steps

1. Specify the design matrix
2. Define the VOIs
3. Enter your chosen model
4. Look at the results
5. Compare models

17
Specify design matrix

• Normal SPM regressors
• -no motion, no attention
• -motion, no attention
• -no motion, attention
• -motion, attention
• DCM analysis regressors
• -no motion (photic)
• -motion
• -attention

18
Defining VOIs

• Single subject choose co-ordinates from
  appropriate contrast.
• e.g. V5 from motion vs. no motion
• RFX DCM performed at 1st level, but define
  group maximum for area of interest, then in
  single subject find nearest local maximum to this
  using the same contrast and a liberal threshold
  (e.g. Plt0.05, uncorrected).

19
specify
NB in order!
20

• Can select
• effects of each condition
• intrinsic connections
• contrast of connections

21
Output
Latent (intrinsic) connectivity (A)
22
Modulation of connections (B)
23
Input (C)
24
Comparing models
See what model best explains the data, e.g.
Original Model Attention modulates V1 to V5
Alternative Model Attention modulates V5
?
25
(No Transcript)