| Author |
 Topic  |
|
|
palekar
Starting Member
USA
1 Posts |
 Posted - Jun 24 2004 : 12:24:22 PM
|
Hi all,
It has been noticed again that the co-registration of the anatomical with the functionals in the SPM analysis of fMRI data simply worsens the results. The mis-registration (error) can be easily seen while observing the SPM t-maps projected onto the glass brain. The whole t-map appears shifted from the glass-brain image.
On the contrary, if we DO NOT carry out the co-registration step, the registration with the glass-brain image looks almost perfect.
I am curious to know if anyone has experienced the same mis-registration problem after performing the SPM coregistration step.
-Parag |
|
|
diaz
BIAC Alum
USA
212 Posts |
Posted - Jun 24 2004 : 1:39:34 PM
|
Hi,
I have heard of, but personally not experienced this. I use tstat profile to analyze my data rather than SPM. Because of the ease with which preprocessing can be done now (bputil - thanks Jimmy), I have actually analyzed my data both ways to examine these potential effects. In comparing t-maps derived from either coregistered and normalized or from just normalized data, (overlaid on the individual subject's normalized brain) there are not substantial differences. Roughly the same patterns of activation can be seen both in individual subject data and in group averages. So I don't coregister. Whether this is the correct choice for you may depend on the specific parameters of your study. |
Michele T. Diaz, Ph.D.
Associate Director
Brain Imaging and Analysis Center |
 |
|
|
melissa.slavin
BIAC Alum
Australia
44 Posts |
Posted - Jun 24 2004 : 1:56:44 PM
|
I am definitely not the SPM expert around here, but given I seem to be the loudest SPM person, I'll have a go at answering this. First, I would check your individual normalisations. If someone had a bad normalise, that can muck up the appearance of the group data. Second, if you have anything but a young, healthy group, you may have to consider creating your own template for normalising to. A third option (which we have recently taken to avoid these problems) is to normalise to the EPI (functional) template, and skip co-registration. You then don't deal with the anatomicals at all. Finally, remember that there is a small discrepancy between MNI space and Talairach space which can cause some funny looking results on the glass brain.
Melissa |
|
|
|
charles.michelich
BIAC Alum
USA
183 Posts |
Posted - Jun 24 2004 : 2:06:54 PM
|
Melissa,
Have you been successful normalizing your functional data against the EPI template? I believe that when Beau wrote some of the early pre-processing scripts, he found that normalizing spiral functional data, particularly spiral in data, failed miserably.
Chuck |
|
|
|
melissa.slavin
BIAC Alum
Australia
44 Posts |
Posted - Jun 24 2004 : 2:19:54 PM
|
I personally am dealing with old data which is not spiral. Our medical student, Sriyesh, has successfully normalised our more recent data which is inverse spiral to the MNI template. He then manually corrects any problems with roll, pitch, yaw etc. He then skips the coregistration step, which assumes that there is not substantial or systematic movement between anatomical and functional runs. We looked at the results of this compared with the alternatives, and this seemed to be the best. It avoids the systematic error that occurs with co-registration of our inverse spiral data at least. If someone has a better suggestion which can also account for motion, I'd really be interested. I think there was some suggestion from the SPM list of creating our own template for inverse spiral images, but that brings up a whole new set of problems. We may look into it once we have enough data.
thanks,
Melissa |
|
|
| |
Topic |
|
|
|
Co-registration in SPM
