fMRI Journal Club

1
Evaluation of mixed effects in event-related fMRI
studies Impact of first-level design and
filtering M. Bianciardi, A. Cerasa, F. Patria,
and G.E. Hagberg Neuroimage 22 (2004) 1351-1370
fMRI Journal Club September 28, 2004 Andy James
and Jason Craggs
2
Problem What is the best design and analysis
approach for event-related fMRI (er-FMRI) studies?
Designs Block Bimodal Geometric Latin
square Bimodal Fixed
Analysis SPM99 SPM2 FSL3.0
We are primarily interested in fixed effects
(factors we control) and not random effects
(factors varying by subject). We want results
that are sensitive and specific.
3
What is an event related design?

• An fMRI experiment where stimuli are presented as
  individual discrete trials, which can vary both
  time and sequence of stimuli
• In contrast, block design experiments typically
  have groups or blocks of trials

Event
Block
4
Different event related fMRI designs
Binomial
Hayberg 2001
5
Fixed, Random, and Mixed

• Fixed effects factors and levels that the
  experimenter is arbitrarily and systematically
  choosing to analyze
• Random effects factors the experimenter is not
  attempting to control, but will use to test
  external validity
• Mixed effects analyses that incorporate fixed
  and random effects

6
Random effects analysis
In performing a standard GLM analysis, the
resulting significant results are strictly
speaking only valid for the group(s) of subjects
or patients included in the analysis because
subjects are treated as a fixed effect in a
standard GLM. In order to generalize the obtained
fMRI results to the population level, a random
effects analysis has to be performed. This means
that the studied sample of subjects are treated
as a random selection from the population of all
people. Note, that for generalization to the
population level, many subjects should be
included, i.e. 50 or more (per experimental
group!). With a few subjects, it is simply
impossible to estimate general population
effects. The recommended minimum for random
effects analysis are 10 subjects per experimental
group.
7
Other Key concepts

• Sphericity
• Is an extension of homogeniety of variance, but
  with a repeated measures twist
• That is, we expect the covariances between
  groups/regions/whatever to be roughly equal
  across multiple measurements

8
Sensitivity

• How well can the hardware detect an fMRI signal
  elicited from the paradigm

9
Precision

• How well can you specify the origin of the
  detected signal.
• How much is it really related to the experimental
  manipulation

10
Anatomic ROI masks

• Red M1 (Active) Green S1 (not active)
• Used for 2nd level (random effects) analyses
• Also depict ROIs for comparing designs and
  programs

11
Sensitivity vs Specificity/Precision