Dear Jesper,

I get the best results when using the T1 template and the structural
image for normalization. Then applying the resulting deformation matrix
on the stuctural images.

Even the process of reorienting the functional images to have a better
overlap with the template leads to similar disappointing results. Additionaly
I heard that this is a common of preceding step needed in most cases
if using SPM99. A colleague told me, that reorienting images is not 
needed with SPM2 anymore. Please correct me if I am wrong.

Thank You for Your help !

Ferenc

------------------------------------------------------------
Ferenc Acs
Lehrstuhl Prof. Dr. M. W. Greenlee
Institut für Psychologie
Universität Regensburg
93040 Regensburg
Tel. +49 (0)941 943 3582
Fax +49 (0)941 943 3233
http://www.psychologie.uni-regensburg.de/Greenlee/team/Acs/acs.html

>>> Jesper Andersson <[log in to unmask]> 11/29/04 10:14  >>>
Dear Ferenc,

I dont think your problem is specifically linked to anything in your 
pre-processing chain. I believe it is simply a consequence of the 
combination of i) Small FOV in z-direction and ii) Poor starting 
estimates for the normalisation. The second of these you can rectify.

The normalisation is iterative and proceeds by taking multiple steps 
along directions that makes the cost-function smaller (i.e. that make 
the template and object images look more similar). If the distance 
between the images is too large to begin with it may start going in the 
wrong direction and get stuck in some local minima. What you need to do 
is to make sure that the origin is placed in _roughly_ the same spot in 
the template and object images before you start the normalisation. Also 
try and make sure that the pitch-angle (the nodding movement angle) is 
roughly the same for the two images.

You can use the SPM "Display" function to set origin in (in terms of all 
size position parameters) the object image. There are many previous 
replies to the mailbase detailing how this should be done. I suggest you 
search the archives with e.g. "Reorient Images" as search term.

>By the way: Then I ask myself why to unwarp functional images at all, if the 
>corrections of EPI field distorsions will harm the normalization process.
>
>  
>
It is true that interpolation is "harmful" to the images in terms of 
their information content, and therefore one should in principle minimse 
the # of interpolations. However, if one does an intial interpolation 
(after "Realign" or "Realign and Unwarp") then the second interpolation 
(associated with spatial normalisation) will affect all the images in 
the time-series in the same way. That means that the second 
interpolation is effectively a smoothing (with a very small smoothing 
kernel) that is common to all images and will be completely swamped by 
any "Smoothing" that is subsequently applied as part of the analysis 
process.

It should also be realised the Unwarp (and "Slice Timing", which will 
also cause an extra interpolation) is there for a reason.

>Seems quite proper, except the 'ghost image' on the lower left functional image.
>
>  
>
The ghost is just an effect of the way you display your images. If you 
switch to "Nearest neighbour" interpolation in Display it will go away.

Good luck Jesper