Dear Serena,
I'm afraid I don't know of any good papers about this kind of
basis function usage. Maybe someone else on the list can
help with this.
From these results it seems that all you have is "deactivation"
with respect to baseline. I would be tempted to run MELODIC
on this data and see if you found any interaction with resting
state networks or other activation-like signals that have been
missed by your model. It is possible that MELODIC might
show you some feature or your data that your model isn't
capturing, or some mis-specification of your model wrt your
task.
It is possible that what you are seeing is biologically sensible
but that would require a good understanding of your task.
I suggest that you discuss this with others more expert in the
area of your task as well.
All the best,
Mark
On 12 Jul 2011, at 11:27, sere f wrote:
> Dear Mark,
>
> I have followed your advice and I have got the below result:
>
> C1= [0 1 1 0 0 0 0 0] empty zstat map
> C2= [0 -1 -1 0 0 0 0 0] empty zstat map
> C3= [-1 0 0 -1 -1 -1 -1 -1] NO empty zstat map
> C4= [1 0 0 1 1 1 1 1] empty zstat map
>
> Considering that
> only FIR 2 and FIR 3 give (very small) activation and the others FIRs give only deactivation,
> do you think that the results of contrasts are reasonable?
> I have a doubt about FIR 3, it shows activation in very small regions and deactivation in other small regions. Do you think it is possible?
>
> Finally, could you suggest me some papers about GLM in event related design using basis functions?
>
>
> Thanks again for your help.
>
> Serena.
>
>
>
>
> On Mon, Jul 11, 2011 at 3:33 PM, Mark Jenkinson <[log in to unmask]> wrote:
> Dear Serena,
>
> The temporal origin of the FIR is at the beginning of
> each stimulus if you have set Phase to zero.
>
> The difference between your C1 and C2 below is just
> that it isn't comparing the second two bins versus
> baseline but versus the average of the other bins.
> So, for example, if you had a reasonable amount of
> undershoot then you might see bigger responses to
> C2 than to C1. You really need to look carefully at
> the results you are getting to know exactly what is
> happening though. It would be interesting to use the
> following contrasts:
> [0 1 1 0 0 0 0 0]
> [0 -1 -1 0 0 0 0 0]
> [-1 0 0 -1 -1 -1 -1 -1]
> [1 0 0 1 1 1 1 1]
> to separate out the various positive and negative
> responses that may be going on versus baseline,
> which will help you understand what you see in your
> C2.
>
> All the best,
> Mark
>
>
> On 11 Jul 2011, at 12:49, sere f wrote:
>
> > Dears Mark and Luca,
> > thanks for the replys. They have been really useful for me.
> >
> > I think i have understood F-Test and how to apply it. It is a logical OR between significativity of activation and deactivation among EVs.
> >
> > I would like to comment a bit the 2 contrasts:
> > C1 [0,1,1,0,0,0,0,0]
> > C2 [1/3 1 1 -1/3 -1/3 -1/3 -1/3 -1/3]
> >
> > I have obtained that the first contrast gives empty zstat map and the second gives zstat map with activation more or less everywhere.
> > This result has surprised me. The EVs are 8 FIR basis functions.
> > It means that EV1 and EV2 explore time intervals consecutive....how is it possible that they are not significative against baseline and they are significative against the other time intervals?
> > is my interpretation of basis function correct?
> >
> > In addition, where is the temporal origin of FIR basis function?
> > Does it correspond to the first event written in the Custom-3Column.txt file?
> >
> > Thanks .
> >
> > Serena.
> >
> >
> >
> >
> > On Mon, Jul 11, 2011 at 9:45 AM, Mark Jenkinson <[log in to unmask]> wrote:
> > Dear Serena,
> >
> > Luca is correct about the final contrast question.
> >
> > As for the F-tests, they always tell you when the
> > combination of one *or* more EVs is significantly
> > different from zero. So in your first case, zfstat1
> > tells you when any combination (one or more)
> > of the EVs is substantially different from zero.
> > You certainly *do not* need them all to be different
> > from zero. It is asking the *or* and not the *and*
> > type of question. So if you get a significant result
> > from zfstat1 then you don't know which of the
> > EVs (or which set of EVs) is non-zero - to do this
> > you need to look at the individual contrasts (C1-C8).
> > Also, the F-test is always unsigned, which means
> > that it responds equally to positive *and negative*
> > changes in the contrasts (unlike the T-tests which only
> > respond to positive changes).
> >
> > So zfstat3 will tell you where either C2 *or* C3
> > *or both* are substantially different from zero.
> > And that could be because they are positive or
> > negative - either way, and in any combination.
> >
> > Finally, zfstat2 is an F-test only on contrast C1
> > (so I assume you mentioned C2 in your question
> > by mistake). If you see a stronger result in zfstat2
> > compared to zstat1 (i.e. in F2 vs C1) then this is
> > most likely to be caused by negative changes
> > which C1 will not pick up (the individual T-tests
> > are signed and so only look for positive changes).
> > Otherwise they use the same information and the
> > F-test threshold is slightly higher than the T-test
> > one since it is two-sided, and so there may be some
> > things that pass the T-test threshold that do not
> > pass the F-test threshold.
> >
> > I hope this answers your questions.
> > All the best,
> > Mark
> >
> >
> >
> >
> >
> >
> > On 10 Jul 2011, at 11:00, sere f wrote:
> >
> > > Hi All,
> > >
> > > I need some advice to understand well the activation maps that I obtain in ER design.
> > > I use 8 FIR basis functions in Real EV setting.
> > > This is the setting for contrasts and F-test.
> > >
> > > F1 F2 F3
> > > C1 10000000 * *
> > > C2 01000000 * *
> > > C3 00100000 * *
> > > C4 00010000 *
> > > C5 00001000 *
> > > C6 00000100 *
> > > C7 00000010 *
> > > C8 00000001 *
> > >
> > > I obtain 8 zstat maps, F1, F2 and F3 zfstat maps.
> > >
> > > Does F1 zfstat map tell where the activation of all 8 functions is significantly different from 0?
> > > F2 zfstat map has a bigger activation then C2 zstat map....is it due by the p-value on C2?
> > > Does F3 zfstat map tell where the activation of C2 and C3 is different from 0?
> > >
> > > In addition, the contrast [0,1,1,0,0,0,0,0] asks about the average response of 2 basis functions versus baseline.
> > > Is the contrast [-0.33,1,1,-0.33, -0.33, -0.33, -0.33, -0.33] different from the previous one?
> > > And what is the info that they give compared F3 zfstat?
> > >
> > > Thanks in advance for any suggestions.
> > >
> > > Serena.
> >
>
|