Hi Vadim
> 1. So, according to what you are saying if I use FIR I have no way to see
> t-contrast based activations in a glass brain (unless they are averaged My
> design is faces / objects, which localizes faces selective regions as
> t-contrast faces-objects. So, I have no use of F-contrast per se. As you
> suggested I can plot the regressors response, but this still doesn't provide
> me with overview of the whole brain.
Right. However, it might not be as bad as you think. For example, if
you have an idea of where you expect a faces > houses difference (i.e.
in the fusiform gyrus), and you run an F test, and you get a
difference in a reasonable anatomical location, you can just check the
response in that region to ensure it's in the expected direction.
At the same time, it's nearly always the case that a t-test on the
canonical HRF will be more straightforward to interpret; if this seems
to work well for your data (and the contrast you are interested in), I
would go with that.
> 2. Does not HRF estimated t-contrast make some sort of average? More
> general, is there any rule of thumb when it's preferable to use FIR instead
> HRF? Would it be correct to say that for Event-Related design the FIR is
> more suitable?
There are several other posts on this; you might also want to take a
look at Chapter 30 of the SPM8 manual (Face group fMRI data), which
does some comparisons of basis functions.
For block designs, the assumed shape of the HRF probably matters less,
because most of what you measure is going to be more sustained, so I
suspect that these considerations are more relevant for event-related
designs.
Whenever you create a first-level model, you are describing how you
think the BOLD signal will be in response to your experimental design.
A canonical HRF presumes a a particular shape to this response, and
because of this is fairly easy to interpret. Of course, for brain
areas that are NOT well described by a canonical HRF, your model will
not be very good. Including derivatives of the canonical HRF will
help with this, though (again, see chapter 30 of the SPM manual).
An FIR model makes no assumptions about the shape of the response.
Thus, it's much more flexible, and well-suited for capturing a wide
variety of responses. The downside is that this flexibility makes it
more complex to interpret.
The question is really, is there reason to think that the HRF in the
regions you are interested in differs significantly from the canonical
HRF (or the canonical HRF + derivatives)? Most of the time these
functions seem to do a pretty good job. However, this isn't often
quantitatively evaluated, and may differ based on region, age,
task...etc.
Hope this helps,
Jonathan
|