When you say smaller, do you mean that the actual dimensions (numbers
of voxels) are smaller, or that the brains in the image volumes have
shrunk. If you mean the volume dimensions, then check the bounding boxes.
If you are referring to the brains within the volume shrinking, then
I have my suspicians about the source of the problem. My first
thought was that SPM was picking up wrong .hdr or .mat information,
but renaming the file and trying different .hdr files didn't fix it,
so I guess that this is not the source.
I think what could be the problem is related to the masking that is done
to define voxels who's values are not known. Floating point format images
have values of NaN (Not a Number, normally generated by 0/0) inserted into
these voxels. There is no NaN representation for most of the other image
data types, so zero is used instead. I am also assuming that you are
using sinc interpolation to resample the images. The interpolation routines
don't cope to well with NaNs, so if you use sinc interp, then any voxels
that are within half a sinc kernel from a NaN will come out as NaN. With
bilinear interpolatin, then only the closest voxels will come out this way.
Good luck,
-John
| Date: Thu, 11 Nov 1999 18:04:32 -0500
| X-Accept-Language: en
| MIME-Version: 1.0
| Content-Transfer-Encoding: 7bit
| Subject: normalizing contrast images in spm99b
| From: [log in to unmask]
| To: [log in to unmask]
| X-List: [log in to unmask]
| X-Unsub: To leave, send text 'leave spm' to [log in to unmask]
| X-List-Unsubscribe: <mailto:[log in to unmask]>
| Content-Transfer-Encoding: 7bit
|
| Hi - I've been trying to normalize contrast images in spm99b by applying
| an sn3d.mat file derived from a raw functional MR image, and have been
| having no luck. The un-normalized MR and contrast images are perfectly
| registered (at least as far as I can tell using Check Reg), and all of
| the dimensions and origins are the same. The only difference is that
| the contrast image is a float whereas the MR image is an int16. I
| derive an sn3d.mat file using the MR image, and then apply it both to
| that image and to the coregistered contrast images. However, the
| normalized contrast images are appreciably smaller than the normalized
| MR image in the Y and Z dimensions - they seem to be rougly equal in the
| X dimension. The problem occurs whether or not the con_0001.mat file is
| present.
|
| The strange thing is that the same process of normalization works just
| fine for all of the other files created by the analysis (mask.img,
| spmT_xxxx.img, beta_xxxx.img), and is specific just to con_xxx images.
| The problem persists even if I copy the contrast file to a new name, and
| if I use a different header file (e.g., one copied from the spmT_xxx
| file that normalized correctly).
|
| The one way that I have been able to fix the problem is to convert the
| contrast image from float to int and create a header using makeaheader
| (from AIR). The resulting file normalizes correctly - however, I would
| rather not have to do this with all of my contrast images.
|
| Any help with this problem would be greatly appreciated.
|
| cheers,
| russ
|
| --
| Russell A. Poldrack, Ph. D.
| MGH-NMR Center
| Building 149, 13th St.
| Charlestown, MA 02129
|
| Phone: 617-726-4060
| FAX: 617-726-7422
| Email: [log in to unmask]
| Web Page: http://www.nmr.mgh.harvard.edu/~poldrack
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|