Event-related designs

1
Event-related designs

• Paul Wright
• Liu lab meeting Feb 1st 2005

2
Why use event designs?

• The alternative block design
• Habituation
• Anticipation
• Need to randomize stimuli
• Analyze trials by performance

3
Review
4
Trial timing
SOA Stimulus Onset Asynchrony
ISI Interstimulus interval
ITI Intertrial interval SOA
5
Dale and Buckner (1997)

• Key concept selective averaging
• Hemodynamic response (HDR) can be separated at 5s
  and 2s ITI
• Activation is robust at 2s ITI
• Fixed ITI
• Two trial types randomized

6
Dale and Buckner (1997)

• Selective averaging
• Signal is averaged relative to event start
• Event types averaged separately
• Activation covariation between
• Observed signal
• Normalized predicted HDR

7
Dale and Buckner (1997)
8
Dale and Buckner (1997)
9
Dale and Buckner (1997)
10
Dale and Buckner (1997)

• Caveats
• We did observe subtle departures from linearity
• results apply only to simple visual
  stimulation
• higher-order task trials may produce
  non-linearity due to the underlying neural
  activity

11
Dale and Buckner (1997)

• the BOLD fMRI responses add roughly linearly,
  even when the trials are spaced as little as 2
  sec apart.

12
Dale (1999)

• Simulated experiments
• Calculated estimation efficiency
• Ability to accurately calculate HDR
• Shorter ISI is better, if variable
• Not all random patterns are equal

13
Dale (1999)
14
Dale (1999)

• A significant improvement typically can be
  achieved by generating a large number of
  candidate experimental designs using a stochastic
  process, and the selecting the one affording the
  greatest estimator efficiency.

15
Soon et al. (2003)

• Used faces presented for 1.5 sec
• Single face or pair
• ITI either 3 sec or 6 sec
• Second face either the same or different
• The second face response was smaller
• With shorter ITI
• With identical faces

16
Soon et al. (2003)
17
Soon et al. (2003)

• Simulations
• Jittered 9 sec ITI (6, 9, or 12 sec)
• Jittered 6 sec ITI (3, 6, or 9 sec)
• 3 sec ITI with null trials (mean ITI 4.5 sec)
• Simulated 20 signal loss with 6 sec ITI, and 40
  with 3 sec ITI
• Detection power increases with ITI

18
Soon et al. (2003)

• Rapid designs suffer from both neuronal and
  hemodynamic apaptation
• Shorter ITI lowers power, but
• The higher variance is compensated by the
  increase in power due to the larger number of
  events.
• Short ITI minimum 3 sec,average 4.5 - 6 sec

19
Serences (2004)

• Simulated experiments
• Compares event-related averaging with
  deconvolution
• Explains how jittering works
• exponential distributions have been shown to
  yield more efficient experimental designs than
  uniform ISI distributions

20
Serences (2004)
The critical constraint on solving any system of
linear equations is that there must be as many
unique equations as there are unknowns.
21
Serences (2004)
22
Serences (2004)
23
Serences (2004)

• Deconvolution is superior to event-related
  averaging when stimuli are sequence dependant
• Is Serences method the same as our GLM?
• When stimuli are random, both methods can
  estimate the HDR

24
Summary

• What is the ideal ITI?
• Dale says as low as 2 sec
• Soon and Serences 3 sec
• How to jitter?
• random conditions (Dale)
• randomized ITI (Soon)
• null trials (Serences used exponential
  distribution)

25
Summary

• How to analyze?
• Dale averaged events first, then fit to a curve
• Brainvoyager creates a predicted time course for
  the entire experiment
• Deconvolution and event-related averaging are
  both acceptable if stimuli are not sequence
  dependant (Serences)
• How many trials?
• Numbers vary from minimum of 25 up to 60 per
  condition

26
What now?

• Jessicas event-related face matching design
• Randomized conditions
• No jitter
• ITI 5 sec

27
What now?

• My go/no-go task
• Null trials (300 ms) shorter than trial (1200 ms)
• Mean ITI 1800 ms

28
What now?

• Effect size from the matching experiment
• Response to faces is usually gt100 of control
  response