hello Cyril
maybe I explained myself badly, or you answered my question perfectly
and I don't quite get your answer ;-)
the thing was to restrict analyses to areas that are relevant for the
verb generation task before comparing groups (not entire brain,
because then differences between groups are too small).
so to define which areas are activated by verb generation (versus
control: -1 0 1 as only contrast that interests me, not the
drawings), I perform a 1ST with all sessions.
so 37 sessions with results from the contrast -1 0 1 (individual,
first level analysis) are used in a 1ST test with contrast 1 (second
level analysis) to assess general verb generation activation.
this gives me a con.img, a mask.img and a SPM_T.img
one of these three can then be applied as mask for other analyses
(anova or 2ST, ...) to assess differences between groups. in this way
I only look at different activations between the groups in areas that
are relevant in a verb generation task per se.
for instance, say I want to know what areas are activated more in L2
then L1 I'll perform an anova (second level) with two groups: I take
my con.img (from first level) of all bilingual kids in L1 as group 1
and of L2 as group 2. when building the model, I then say to mask it
with the "main effect" mask from my 1ST test of all sessions (at the
"explicitely mask images?" prompt.
the question is then what image (from the 1ST of all sessions) to use
as mask image at this stage.
- mask.img
not the mask.img you say, but isn't that precisely what i want. it it
is a binary image, i guess the 1's stand for "activation" (although
not necessarily significant" and the 0's stand for "no activation".
or am i wrong? because applying this as a mask would exclude looking
at areas that are not activated during verb generation. so only
looking at "relevant" areas.
But then the question remains (if I am correct about the activated =
1 and not activated = 0) at what height threshold this assigning of
1 and 0 comes about? and how can I manipulate this threshold (because
as it is, it is practically the entire brain)
- SPM_T image
If i take the one SPM_T image generated from my 1ST test (contrast
1) of all sessions and in ImCalc i say to calculate the image based
on activations exceeding a height threshold of 3 (for example) . So
Imcalc, choose 1 image: SPM_T, output: main-effect.img, evaluated
function = i1>3
do I not get then a proper mask, ie an image where only areas are
included that are activated (not necesarily significant) during verb
generation in all sessions with a controlled height threshold of 3,
which i can use for the intergroup analyses? (the image in attachment
main-effect.img.hdr is the results of this operation)
-con.img
or can i use the con.img, because when using check-reg in spm and
displaying my self-made main-effect mask (from the SPM_T via imcalc)
and the con.img, white regions in con.img = white regions in the
main-effect.img (which is, of course, obvious). But then what are the
values in the con.img precisely and how does SPM decide which "grey"
is included in the analysis and what "darker grey" is excluded? and
how could I manipulate that?
i hope i'm not driving you crazy ;-)
thanks
katrien
>Hi katrien,
>
>>Hi
>>I am working on fMRI data of bilingual children. I want to do
>>comparisons between groups (one way anova's) Presently I am working
>>on a verb generation from a block design, 3 trials: drawings (1),
>>nonsense words (2), and verb generation from visual noun (3) .
>>contrast of interest here is (-1 0 1) on individual level.
>> I read that some people use a main effect mask when building the
>>model when performing group comparisons. Indeed it sounds clever
>>to only include differences in activation from the activations seen
>>in the overall group/ the task. Therefore, the idea is to produce
>>the effect of the whole group (bilingual and monolingual, L1 and
>>L2, ...), which is then used as a "main effect" mask for later
>>inter group analysis.
>
>Well, one simple possibility here is to perform your (-1 0 1)
>contrast for each subject and then oppose monolingual and bilingual
>children with a t-test (an anova with 4 groups: monoling drawings,
>monoling nouns, biling drawings and biling nouns; will mix within
>and between variances...)
>
>>So, I started playing around and found in the list that you can use
>>the con.img of the relevant contrast of all sessions, main effect,
>>as mask. although there is also a mask.img file or I could use the
>>SPM_T map and calculate (via Imcalc) for instance i1>3 to include
>>only areas with activations that are higher than 3
>
>I guess you performed two one-sample t-tests monoling drawings and
>monoling nouns that gave you two spmT maps (or two two-sample
>t-tests mono>biling drawings and nouns). The spmT are only the ratio
>con.img/variance and doing i1>3 with imcalc will NOT give you areas
>*significantly* more activated for i1 in comparison with i3. If you
>want such a result, you have to do it with a test.
>
>>1- is it best to use the SPM_T map calculated at threshold >3 (for
>>instance)?
>>2- can i use the con.img? because that are not 1 and 0 values but
>>numerical data, so what am I really doing then? (because it does
>>work)
>>3- do i simply use the mask.img generated after performing the
>>single group analysis of all sessions? if yes, what is then the
>>threshold used to generate the mask.img? and can I change that and
>>how?
>>4- all group comparisons are of course RFX analyses. should i use
>>the RFX 1 sampled T test of all sessions as base for main effect or
>>can I use the FXD analysis?
>
>The mask.img should not be used as a mask for your analysis. It
>corresponds to the common image of all volumes of your subject. It's
>a binary image used to determine which voxels have to be analyzed.
>I would say that the best way is to run your anova and look at the
>'main effect' that gives you where, at least one condition shows an
>effect. Then, when doing a contrast between conditions/groups,
>specify inclusive masking and select 'main effect'. Another option
>is to save a given result as an image (a 'main' effect) and then use
>it.
>
>
>Hope it helps,
>Best
>-cyril
--
_______________________________________
Katrien Mondt
Prospective Research for Brussels
BHG - RBC
Vakgroep Taal- en Letterkunde
Faculteit Letteren & Wijsbegeerte
Vrije Universiteit Brussel
Pleinlaan 2
1050 Brussel
tel: 00 32 2 629 2668
fax: 00 32 2 629 3684
[log in to unmask]
www.vub.ac.be/DITO
www.vub.ac.be/SGER
_______________________________________
|