Hi Philipp, thank you for your comments, and sorry for being unclear about the processing stream. Actually, I do NOT perform any segmentation before normalisation (it is just a *_sn.mat file, not a *_seg_sn.mat file), since I am analyzing data from small rodents, for which I do not have tissue probability maps (and no experience so far as to which tissue types would be suitable for segmentation, anyway). So -- are there any other ideas? Thanks again, Andreas ---------------------- Biozentrum of the University of Basel Klingelbergstrasse 50/70 4056 Basel Switzerland ========================================================================= On Tue, 2 Jan 2007 17:47:53 +0100, Saemann Philipp <[log in to unmask]> wrote: >Hello Andreas, > >I do not know if this effects would appeaar if you use the combined >segmentation/normalisation routine from SPM5 (which gives you *seg_sn.mat >files). >It was not entirely clear from your mail how you have gained the *sn.mat >files (i. e from >the segmentation or the normalisation routine). > >If it is *seg_sn.mat" files you may try the VBM toolbox function "write >segmentation only", >which is the equivalent to "write only" of earlier SPM versions or the SPM5 >normalize function. > >Download from: > >http://dbm.neuro.uni-jena.de/vbm > >However, programmers may have additional ideas on your question. > >best regards, >Philipp > > >Max-Planck-Institute of Psychiatry >NMR Research Group >Kraepelinsr. 2-10 >80804 Munich >Mail: [log in to unmask] >Phone: 0049-89-30622-413 >========================================================================= On Tue, 2 Jan 2007 16:22:45 +0000, Andreas Bruns <[log in to unmask]> wrote: >Dear SPMers, > >applying a *_sn.mat file to an image involves explicit resampling by means >of the "Normalise: Write" option or the "Deformations" utility. When using >nan(1,3) for the voxel size and nan(2,3) for the bounding box, the >dimensions of the template image are used, BUT with a slight tolerance, so >that in general, there is no exact one-to-one (voxel-by-voxel) mapping >between the template and the normalised image. Instead, I repeatedly >obtained a 1-voxel discrepancy in image size and/or a half-voxel offset >between the images in one or more direction(s). > >Certainly, these differences are well accounted for by the information in >the header file, so that the images are correctly co-registered. Still, if I >want to have perfect spatial correspondence between the images, I need an >additional resampling step for the normalised image by means of, e.g., the >"ImCalc" feature. This requires about twice as much time and disk space, and >(what is worse) it introduces more smoothing due to the additional >interpolation step. > >Is there a way to avoid the double resampling and directly tell the >"Normalise: Write" or "Deformation" (or any other) module(s) to use not only >the same voxel size and bounding box as in the template (or even better: in >any arbitrary) image, but even the very same voxel positions and image >dimensions? > >Thank you very much in advance, > >Andreas >========================================================================