For many fMRI studies, the limitation in using your second method is
the coregistration between the fMRI and anatomical data. Distortions
preclude accurate rigid-body alignment, which means that normalising
transforms determined from the anatomical data may not work so well
for the functional data.
Because it is based on the mean squared difference, the first method
is dependent on the image contrast and artifacts in your data matching
those in the template data. If these differ, then the alignment may
not be so good.
Often, it is possible to use the spatial normalisation transform
obtained by segmenting the fMRI data themselves. This is less
dependent on the image contrast of your data, and it does not rely on
an accurate alignment with structural data.
Whenever you do an F contrast, you are essentially doing a model
comparison at each voxel. Many other aspects of SPM are also based on
model comparisons. Spatial normalisation also involves fitting a model
to the data, so sometimes it is useful to compare these to see which
is a better fit for your particular data.
Best regards,
-John
On 29 July 2011 15:58, Ce Mo <[log in to unmask]> wrote:
> Dear SPMers,
>
> I ran into a very confusing problem when I tried to analyze my data in 2 ways of normalisation.
>
> The first method is probably a more commonly used one, Normalisation Estimate&Write, which directly transforms the EPI data to the standard MNI space. However, I do not get very ideal results using this method. The extension of voxels survived FWE of the same cluster is too small, (only 2 or 3 voxels in one cluster). And I was hoping to get clusters of bigger size. Therefore, I tried to use an indirect way of normalisation.
>
> First I performed corregistration, using high resolution T1 weighted image(structural image) as source image and mean EPI image generated in realignment as reference image. Then I performed segmentation. And finally I performed normalisation: write. The underlying idea is to first map the EPI data to the structural image, then transform it into MNI space. And the reason for this is that T1 weighted image contains more information, which I hope would lead to better results.
>
> However, the results are even more terrifying!! I got none activation using the 2nd method!!! Would you please help me with my problem? I am actually under a deadline... The batch using the 2nd method I used is attached in the mail.
>
> Thank you all in advance. Any suggestions will be greatly appreciated!!
>
> Ce
>
|