 |
 |
Hi again, Sorry (again). I should have followed the thread a bit more closely before sending out my previous answer. I'd forgotten about your previous enquiries (I send up to about 35 emails a day), so the email I just sent you relates to spatially normalising to the SPM templates. I think the amount of smoothing you use should partly depend on how accurate you believe the registration can be. In principle, it should be possible to achieve better intra-subject registration than inter-subject, so you should be able to use less smoothing. The slightly different contrast in the images could be a problem though, as the least squares objective function assumes that one image looks like a spatially transformed version of the other - but with Gaussian noise added. This means that registration of diffusion weighted images may produce slightly biased results (as they don't really conform to the model). I therefore wouldn't really be able to say how accurately they could be registered. Jesper Andersson had a paper at HBM a few years ago about getting diffusion images in register. I think that you may be able to achieve better registration this way. Unfortunately, there is no toolbox in SPM that will do this though. If you want to just try rigid-registration, then the coregister button may do a reasonable job though. All the best, -John > thanks alot. Know it works: wB=B > > But did I need such high smooth factor (8mm)? > I' am not sure, which are the best parameters for my task. > > > The task: > I ' am trying to normalize all diffusion weighted images to the B=0 image > with an Affine transformation. > Could you help? > > ByeBye > > John Ashburner wrote: > > Note that by default, the template is assumed to be smoothed by 8mm. In > > order to match another image to the template, SPM will therefore smooth > > this other image by 8mm => giving you an affine transform that is not the > > identity. > > > > Try changing from > > defaults.normalise.estimate.smosrc = 8; > > defaults.normalise.estimate.smoref = 0; > > to > > defaults.normalise.estimate.smosrc = 8; > > defaults.normalise.estimate.smoref = 8; > > > > As an aside, if you are normalising an image to some other image that is > > not an SPM template, then (I think) you can change the voxel sizes to > > [NaN NaN NaN], and the bounding box to ones(2,3)*NaN , and it will write > > out the images with the appropriate dimensions. > > > > Best regards, > > -John > > > > > if I normalize an image B_0 (B_0 source image and image to write) with > > > an affine transformation to itself (B_0 template), I obtain another > > > image than before: > > > wB_0 =! B_0 > > > Does anybody know why? > > > > > > The defaults are modified in the following sence: > > > > > > defaults.normalise.write.vox = [0.89 0.89 3.6]; > > > defaults.normalise.write.bb = [-113 -113 -62; 114 114 63]; > > > > > > and only an affine transformation is allowed. Everything else is not > > > changed. > > -- > Siawoosh Mohammadi > > Dept. of Neurology > University of Muenster Phone: +49-251-8352061 > Albert-Schweitzer-Strasse 33 Fax: +49-251-8348181 > 48129 Muenster, Germany E-mail: [log in to unmask]