 |
 |
Alternatively, you could chop out the parts of the scan where the data is not so reliable. The attached function may help here. Using the Display button, you can figure out the voxel indices of the corners of a subvolume containing reliable information, and then pass these ranges as the second argument. % V = spm_vol(spm_select(1,'image')); % subvol(V,[32 64 ; 1 64 ; 1 48]'); Best regards, -John On 3 January 2013 21:00, H. Nebl <[log in to unmask]> wrote: > Hello Ethan, > > > sorry to intervene, but as I had a similar problem a while ago I also have > to comment ;-) > > I don't think signal inhomogenities are the main reason in your particular > case. Instead it seems you did not cover the whole brain (at least if the > right image shows your original data). Parts of the skull at the right and > left side are missing in your T1 volume, whereas SPM's T1 template includes > skull at these positions. Now the normalization routine seems to simply > stretch the temporal parts to compensate for the "missing" skull. At least > this was my experience with similar data. > > One solution that worked fine (or at least much better) with my data sets, > and here we come to the very same that Marko proposed, was to use > segmentation routines for normalization. As the skull is removed in one of > the segmentation steps and with the different probability maps for GM, WM > and CSF the segmentation algorithm seems to be less prone to missing skull. > So I'd say give it a try with standard parameters first, maybe this already > solves your problem. > > Remember, you have to use the same normalization procedure for all the > data (so either T1 template normalizaiton for all the data or > segmentation), because they will result in slightly different global brain > forms. > > > Best, > > Helmut >