 |
 |
Hi Hojat >> Thanks, I know how the warps that are estimated during the >> segmentation process are written out. The thing I cam not understand >> is why the deformation fields which are used to >> warp probability images inorder to register them with segmented >> parts can be used to normalize the whole fMRI images. Well, the "segmented parts" includes GM, WM and CSF, which together practically cover the whole brain (with typical resolution/contrast you don't need to worry about things like blood vessels, meninges or sinuses being different classes). So the inverse of a transformation that warps the prob maps from MNI to subject, will be a pretty good warping from subject to MNI. > The inverse of this, simply speaking (as far as I understood John, it is > not quite that simple :) will consequently deform the original image to > match the prior in normalized space (i.e., normalize the image). Conceptually, I think it is that simple! In practice, the complication is that inverting a non-affine transformation is quite tricky, I think there is some stuff about this in chapter 3 or 4 of HBF2: http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/ But the basic idea is that if you know where points in MNI space map to corresponding points in subject space, you can work out where points in subject space map to in MNI, and hence normalise subjects to MNI. Best, Ged.