Hello,
thanks for your reply...
I actually did find an error in the registration part of the analysis.
I was moving files from one computer to another and then the links in
*.feat/reg folders got broken. I wrote a script to change them, but I
did not doublecheck the hires files for each subject and I linked some
to the wrong images. That's why the background image seemed "wrong".
After I fixed that, the registration was fine, but the statistical
results remained "different". Now I had no "activations" above the
threshold in the group analysis. When comparing "old" (I am referring
to the one before the current version of feat: I believe that was 3.0?)
to corresponding "new" zstat files, I noticed that the results are not
"completely different" (as I mentioned earlier and as one of the
questions you pointed to states). Instead, it seems that the
distribution of "activations" is more or less the same, except there is
less of it and the z values are consistently lower. For example, the
old zstat file that had a max/min values around 5/-5, the new one
showed them to be around 2.5/-2.5. You can see an example of that at
http://nina.bmap.ucla.edu/ That is quite a big difference! But based on
the links you sent, I guess that is not unusual?
thanks
zrinka
On Jun 25, 2004, at 7:49 AM, Stephen Smith wrote:
> Hi Zrinka,
>
> Wrt registrations and background image, yes if you are only re-running
> higher-level analysis then the old registrations will have gotten used
> and
> not rerun. The background image may have changed because I think we
> added
> in between the current and previous versions a flag to normalise the
> background images to all have the same mean intensity before averaging
> -
> just a display nicety, really.
>
> Wrt the stats, which "old" version are you comparing with? If it's 3.0
> then maybe this is answered at:
> http://www.jiscmail.ac.uk/cgi-bin/webadmin?
> A2=ind0310&L=fsl&P=R5762&D=0&H=0&I=-1&O=T&T=0
> and
> http://www.jiscmail.ac.uk/cgi-bin/webadmin?
> A2=ind0403&L=fsl&P=R517&D=0&H=0&I=-1&O=T&T=0
> ?
>
> Cheers, Steve.
>
>
> On Wed, 23 Jun 2004, Zrinka Bilusic wrote:
>
>> Hello,
>>
>> recently I decided to rerun group analysis on some old data. The
>> reason
>> was that the original group analysis was done using the old version of
>> Feat, and I wanted to see the additional reports from the new version
>> of Feat (the Summary of low-level registration and masks). So I loaded
>> the design.fsf into Feat, changed the output directory, verified that
>> everything is OK, saved and run. But the (statistical) results are
>> different. No only in the report, but I also subtracted zstats images
>> -
>> and there are big differences. I inspected the design.fsf files, and
>> the only difference I could find (apart from the obvious differences
>> in
>> the layout for new and the old versions) is that in the new design.fsf
>> file, the degrees of freedom to standard image was set to 7, whereas
>> in
>> the old one it was 12. But that should not cause any problems because
>> the registrations were already calculated in the single subject
>> analyses, and the first step of the group analysis just applies those
>> transformations - right?
>>
>> Also, I notices that the background image for the two analysis looks
>> quite different - even though in both design files it was set to the
>> "mean highres". I looked at the log files and the commands executed to
>> create those seem to be the same.
>>
>> Am I missing something? The statistical results should not be
>> different? Is there anything else I should be checking? by the way,
>> the
>> single subject analyses were not changed after running the initial
>> group anlaysis...
>>
>> thanks
>>
>> zrinka
>>
>
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
>
|