Bernhard -
> We are running an event related fMRI experiment with a single event type
> comparing a group of 8 healthy volunteers to 8 patients. (TR
> 506ms, 6 slices). Each subjects undergoes 3 fMRI scans (590 scans in each
> run) patients get 3 additional scans after application of a specific drug.
> We want to compare the activation between both groups and within the group
> of patients before and after medication applying a second level analysis.
> First we built a seperate design matrix with the simple hemodynamic
> response function and its temporal derivative as basis functions including
> 3 scans per subject resulting in a design matrix with 9 columns. (2
> seperate design matrices for patients before and after medication) Then we
> calculated the contrast images for the contrast 1 1 1 1 1 1 0 0 0. Those
> contrast images were entered into the second level applying a one sample
> t-Test (patients only, healthy only), a two sample t-test (patients vs
> normals) and a paired t-test (patients after vs before medication).
> The results for the within group analysis show little statistical
> significance whereas single subject analyses show consistent activation in
> very similar regions)
> Our questions are:
> 1.) Is the choice of the basis functions and the contrast definition
> appropriate.
No, it doesn't make much sense to average the parameter estimates for
the canonical and its temporal derivative as you have done in your
1 1 1 1 1 1 contrast ...
> Is it possible to enter a contrast image based on two basis
> functions into the rfx-models or have the two functions to be considered
> seperately (contrasts 1 0 1 0 1 0 or 0 1 0 1 0 1 respectively).
... rather, the former contrast of 1 0 1 0 1 0 is appropriate for
inferences about differences in response "magnitude". The latter
contrast for the derivatives should not be used, because the parameter
estimates for the derivatives cannot really be interpreted in isolation
of those for the canonical. Though certain linear combinations of
parameter
estimates of multiple basis functions may be interpretable, it generally
only makes sense to consider more than one basis function within the
context of a multivariate analysis.
> Do we get
> different results if only the hrf is enterd into the design matrix
> (contrast 1 1 1) (in comparison to a contrast 1 0 1 0 1 0).
The temporal derivative basis function is explicitly orthogonalised
with respect to the canonical, so its inclusion or exclusion would not
"normally" affect the parameter estimates for the canonical ("normally"
implying a reasonable sampling rate, which is fine for your short TR
of 0.5s, because an undersampling of the model can introduce a
correlation between the covariates for the two basis functions, in
which case including/excluding the derivative would have small
effects on the results).
> 2.) Based on an a priori hypothesis what threshold has to be applied in
> the second level analysis?
The same rationale applies as in your first-level, fixed effects models.
You could correct for the volume of your region of interest for example
(using SVC in SPM99).
> 3.) Since addition of more subjects results in more degrees of freedom,
> statistical significance should increase. Nevertheless in our analysis,
> significance seems to decrease.
The degrees of freedom in the error term in the 2nd-level group models
depend on the number of subjects, whereas those in the single-subject
1st-level models depend on the number of scans* (minus correction for
any temporal autocorrelation/smoothing), so the significance generally
decreases in 2nd-level random effects models (given that one usually
has more scans than subjects, and interscan variability is usually
small compared with intersubject variability). This is the price you
pay for a more generalisable inference.
* Note: our use of "scans" refers to a volume acquistion; I think you
are refering to "sessions" earlier in your email.
Rik
---------------------------8-{)}-------------------------
DR R HENSON EMAIL [log in to unmask]
Wellcome Department of
Cognitive Neurology TEL (work1) +44 207 833 7483
12 Queen Square TEL (work2) +44 207 833 7472
London, WC1N 3BG FAX +44 207 813 1420
URL: http://www.psychol.ucl.ac.uk/rik.henson/index.html
---------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|