Thank you very much, Jonathan!
I realize that a beta or a con for 1 time bin are equivalent, but I was looking at how to export several time bins to the rfx simultaneously. What I was aiming at doing was to compare the HDresponses (corresponding 14sec windows) for the 2 groups, so would it be correct to create a con image at the individual level: 11111111-1-1-1-1-1-1-1 and then export those to the 2nd level? Ideally, the best analysis would be a cross-correlation with sliding window but unfortunately I do not know of any way to do this in SPM. It seems that Brain Voyager has a feature that allows that, but I have not found anything equivalent in SPM. Do you? (or anyone else?)
Kind regards,
Olga
----- Original Message -----
From: "Jonathan Peelle" <[log in to unmask]>
To: "Olga Prilipko" <[log in to unmask]>
Cc: [log in to unmask]
Sent: Wednesday, April 20, 2011 2:27:43 PM
Subject: Re: [SPM] FIR rfx analysis
Dear Olga,
This is a bit of a delayed response to your question, but I hope still helpful.
> I would like to conduct a group analysis on data convolved with an FIR basis set. After reading SPM8 manual chapter 30 and the SPM archives though I am unclear on what files to import at the 2nd level - from what I read on the SPM list it seems that it should be beta files for each time bin, but SPM8 manual refers to .con images of time bins. Which should it be?
A con* image is just a linear combination of beta* images. By
convention people generally refer to con* images on the second level.
Let's say you had just 3 time bins; in this case, beta_0001,
beta_0002, and beta_0003 would be the parameter estimates for each
time bin. If you did a contrast like so:
1 0 0
This would give you a con* image that is a linear combination of those
beta images. In this case, it would be:
(1*beta_0001) + (0*beta_0002) + (0*beta_0003)
Thus, you would get a con* image that is equivalent to beta_0001. You
can test this in ImCalc by selecting the beta_0001 and con* image (i1
and i2), and then confirming that the expression i1==i2 gives you all
1s (because they are all equal), or equivalently that i1-i2 gives you
all 0s.
This is a long way around of explaining the relationship, but if you
understand the above then it's easy to see why people can often use
beta images and contrast images somewhat interchangeably.
> My second question concerns the contrast specification for the FIR: I would like to compare the first half of the epoch to the second using a t-constrast - what would be the best way to do this while having an odd number of time bins (15)? I could specify 14 time bins instead, but this would make a time bin a bit longer than my TR - so I would not be looking at the same volume in each time bin, so it seems like that would be a problem?
The short answer is that I would suggest you just ignore your last
time bin and compare the first 7 bins to the second 7 bins (unless you
have good reason to think something very interesting is happening in
that last bin).
However, it might also be worth considering what you are comparing.
For example, if your TR is 2 seconds, 14 bins = 28 seconds. Given
something like a normal hemodynamic response, it seems obvious that
the first half of the bins will contain nearly any response. But, of
course, this depends on your particular paradigm and questions you
want to ask, so there is really no standard rule.
Hope this helps!
Best regards,
Jonathan
--
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|