Event-related fMRI Contrast When Using Constant Interstimulus Interval: Theory and Experiment

1
Event-related fMRI Contrast When Using Constant
Interstimulus Interval Theory and Experiment

• Peter A. Bandettini Robert W. Cox
• Steve Smith Psychology 670 Oct. 22, 2002

2
OR The Peter Bandettini Event-Related fMRI
Cookbook Constant ISI Version
3

• Background Information
• Theory (translated from the original Greek)
• Method of testing the theory
• Analysis
• Applying the Cookbook to our own Research (2
  examples)

4
PART I Background Information
5
Optimal Designs

• Optimal designs are those that yield the largest
  estimated magnitudes with the best statistical
  properties while satisfying the behavioral
  constraints of the experiment (Ollinger et al.,
  2001b)
• low variance
• equal variance across effects
• minimum correlation among effects

6
Stimulation protocols in fMRI
Slide ruthlessly stolen from previous lecture
7
ER vs. Blocked Designs

• Better randomization of task types in a time
  series
• Allows for selective analysis of response data
• particular stimuli
• errors (and the accompanying Oh Shit! response)
• Easier separation of motion artifacts (you can,
  in theory, look at particular trials)

8
Methodological Variables

• Stimulus Duration (SD)
• Interstimulus Interval (ISI)

SD
SD
SD
ISI
ISI

• Altering either SD or ISI alters the response
  function.

9
Methodological Variables

• Stimulus Duration (SD)
• Interstimulus Interval (ISI)

SD
SD
SD
ISI
ISI

• Altering either SD or ISI alters the response
  function.

10
Different ISI Patterns

• Constant
• (relatively) easy to analyze b/c they involve
  simple binning and averaging.
• Does not require the assumption of linearity
• Randomized (Mikes presentation)
• more time efficient
• allow for shorter ISIs

ISI
ISI
ISI
ISI
11
Two Critical Questions

• How does the statistical power of ER-fMRI compare
  to that of blocked designs?
• What is the optimal ISI for a given SD?

12
Two Critical Questions

• How does the statistical power of ER-fMRI compare
  to that of blocked designs?
• What is the optimal ISI for a given SD?

13

• Trade-off Number of trials per unit time vs. the
  degree of attenuation of the hemodynamic signal
  that occurs with close temporal spacing of trials.

14
Three Components of a Signal 1) pre-undershoot
(approx. 2 sec) 2) signal (approx. 6-9 sec to
plateau) 3) post-undershoot (approx. 3 sec)
15

• Signal attenuation or clipping
• If one trial begins before the hemodynamic
  response function has settled back to baseline,
  the two functions (trial 1 and 2) will interfere
  with each other.

A
B
16
Thus...

• The purpose of this paper is to determine the
  optimal ISI for a given SD in a constant-ISI
  ER-fMRI design.

17
PART II The Theory (or What I Understood of It)
18
Bandettinis Goal

• Create a theoretical response function for
  constant-ISI ER-fMRI based on fancy-schmancy
  math.
• Compare theoretical response function to
  experimental data.

19
The Theory (as I understand it)

• We want to estimate the activation in each voxel.
• The catch we dont know the response or the
  baseline level of activation.
• Use matrix algebra magic to get estimators of
  response and baseline activation.

20
The Theory (as I understand it)

• If the stimuli are far enough apart (i.e., the
  signals of each activation do not overlap), then
  we can accurately predict a response function.
• If there is overlap, we get more intimidating
  Greek symbols.
• Thus, we want to find a value that gives us a
  usable function rather than menacing symbols.

21
PART III Method of Testing the Theory
22
Participants

• 5 people (probably Bandettinis family)
• data from 2 were lost due to motion artifacts.

23
Two Tasks

• Passive viewing of an 8-Hz red square (presented
  through goggles)
• Bilateral finger tapping
• Tasks performed simultaneously (hmmm.)

24
Different ISIs
ISI (sec) SD (sec) of Cycles 20 20 924 2 13
20 2 1616 2 2012 2 2510 2 308 2 36
6 2 454 2 602 2 90
Separate time series were run for 9 different
ER-fMRI ISIs. One blocked time series was run
for comparison.
25
Image Acquisition

• Two axial imaging planes (visual and motor
  cortex)
• Echoplanar imaging
• TR 1 sec
• TE 40 msec
• Time series length 360 images

3 x 3 x 7 63 mm3
non-isotropic
From last lectureIn general, larger voxels buy
you more SNR, EXCEPT when the activated region
does not fill the voxel (partial voluming)
26
Hmmm...

• What is gained by having the visual and motor
  stimulation simultaneous?
• Will this pattern generalize to other areas?
• Simple tasks (necessary, as this is a pilot
  study). Can we use this cookbook for more
  complex recipes?

27
PART IV Analysis
28
Image Construction

• Based ROI on data from blocked study.
• Created average plots for each time series
• Created a reference function (just a function in
  which the average function repeats over and over
  again.

29
Image Construction Contd

• Created a correlation image (this is when you
  compare the obtained data to the average data)
• Divided this image by the residual time series
  standard deviation for each voxel in order to
  create a functional contrast-to-noise image

30
Task validity Visual and motor areas were found
to be activated by the tasks.
31
At ISIs of 8 sec or less, the responses are
blunted (over-lapping hemodynamic
functions. Ideal ISI approx. 10-12 sec (similar
pattern to blocked)
32
The cleanest response function is found for
ISIs of 10 and 12 sec (followed by 8). The rest
suck.
33
ISI-10 and ISI-12 lead to images that are similar
to blocked images in resolution.
34
Blocked vs. Optimal ER

• The experimental contrast per unit time for
  ISI-12sec is only 35 lower than that of blocked
  designs.
• For ISI-12sec, the stimulus is on for 14 of
  the time, whereas for blocked, the stimulus is
  on for 50 of the time.

35
In a simulation, Bandettinis model produced data
very similar to that found in the experiment.
36
(No Transcript)
37
The theoretical model produced a similar pattern,
but peaked earlier. (Needs to account for
post-activation undershoot.)
38
PART V Applying the Logic to Own Our Studies
39
Sledge Hammer or Whipped Cream?

• Pilot Study - blocked or constant-ISI ER
• Test - depends on the question
• Whipped Cream study - randomized-ISI ER

40
Before using constant ISIs, ask yourself What
phenomenon are we looking at? What subject
population are we using? Will this give us the
most bang-for-the-buck? Constant-ISI
event-related fMRI is a useful tool in specific
situations.
41

• What would you use this sort of design to study?
  Could you apply it to your own research?
• What patient populations should and should not be
  tested this way?
• The constant-ISI generally shows that the
  hemodynamic response is slightly nonlinear.
  Since the randomized ISI design assumes
  linearity, should we be concerned?

42
Example 1 Expectation of Pain
43
(No Transcript)
44
(No Transcript)
45
Expectation of Pain

• What areas of the brain light up during (1)
  pain and (2) the expectation of pain?
• Pain induced through a balloon that is inflated
  in ones esophagus.
• Nasal intubation

46
(No Transcript)
47
(No Transcript)
48
Three Types of Trials

• Pain trials vs. Pleasure trials vs. No sensation
• Pain the balloon in the esophagus is inflated
  to a pre-determined threshold of pain
• Pleasure a puff of air on the wrist
• No sensation duh

49
Details

• SD 4 sec of conditioned stimulus 4 sec of
  pain/pleasure/nothing.
• ISI 16 sec
• TR 1 sec
• TE 40 msec
• (Dont remember slice s, flip angle, etc).

50

51

52

53
Results

• Pain activated the anterior cingulate and
  somatosensory areas.
• The expectation of pain also activated these
  areas.

54
Subtle Transition to Mikes Presentation

• It makes sense to study pain perception/expectatio
  n using a constant ISI.
• You dont need many trials
• Methodologically difficult to present pain over
  and over again without habituation, violence, etc.

55
Subtle Transition to Mikes Presentation

• But, what if youre interested in something like
  working memory? Or low-level visual perception?
  Or language processing?
• Is there a way to have shorter ISIs, thus
  allowing you to maximize your scanner time???