Dear Serge,
>Dear SPM Users,
>
>In our analysis of fMRI data with simple visual stimulation (light flashes)
>we observed the following:
>using spm96 with global normalization we unexpectedly see large regions
>outside V1 with a signal decrease during stimulation. Most of these signal
>decreases disappear when we do not apply global normalization. It seems that
>the signal intensity of the images obtained during visual stimulation is
>scaled down due to the large signal increases in V1, resulting in artificial
>signal decrease in areas outside V1.
As I wrote in a previous mail this is an known problem.
>However, when using spm97, these areas
>with a signal decrease persist, even when global normalization is not
>applied. We see the same effects of unexpected signal decrease with other
>paradigms too.
>
>Can anyone explain what causes this difference or how to solve it using
>spm97?
Following your mail where you told me that spm97 refers to spm96ER, i.e.
the patch that allowed for analysis of event related studies, I have been
able to look into your question. Your description of the problem strongly
suggested a programming error, but after having scrutinised the code that
does the global normalisation I have been unable to find anything. From
there I set up an analysis of one of our epoch related visual activation
datasets using a boxcar, with and without proportional scaling for spm96
and spm96ER, and I found identical results for both analyses.
Note that in order to precisely mimic an spm96 study in spm96ER one needs
to do some additional tweaking since spm96ER introduced, amongst other
things, convolution of the box-car with the HRF. However, having done this
tweaking (which does not affect any of the code involved in global
normalisation) the results are identical.
One difference that comes to mind (although I am sure that this is not the
explanation) is in the default display at the end of the analysis where
spm96 displayed all voxels above the F-threshold that also had z-scores
above 2.33 AND that belonged to clusters whose size was equal to or above
the expectation of the cluster size. spm96ER did not do this final
exclusion which means that the immidiate output of spm96ER gives an
apparently larger no. of superthreshold voxels. Obviously this is of no
practical consequence, but may at a first glance indicate that there are
differences.
Being unable to replicate your observation I would be grateful for a more
detailed description of the exact design you used. I would also be
interested to know if anyone else has seen this problem.
>I would greatly appreciate any help.
>
>Thank you,
>
Good luck Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|