On Mon, Nov 19, 2012 at 3:59 AM, Janosch <[log in to unmask]> wrote:
> Dear SPM experts,
>
> I would like to do a tensor based morphometry study using ANTS (template
> generation -> registration of individual T1s to template -> calculation of
> (log)jacobian images -> smoothing) and do statistics (correlation with
> behavioral data) on the resulting (log)jacobian images in template space
> with SPM.
>
> I already tried it using the (log)jacobian images as the input images in the
> "estimate model" step as I would do with smoothed grey matter probability
> maps in a VBM study. It works, but the estimation process is quite slow
> compared to e.g. VBM data.
This is fine and it should work. My guess is that the difference is
the size of the images and the lack of using a mask. How many voxels
are in each image and what are the dimensions?
>
> Is this approach valid or does SPM only work correctly with images generated
> by itself or only on images in MNI space? My understanding was that SPM
> should just carry out voxel-wise statistics on any input images, so it
> shouldn't matter if I put in grey matter probability maps, FA maps, con-Maps
> or Jacobian images and if they are in MNI or native space. Is that right or
> did I miss anything?
The group models in SPM are GLMs. As such, it can be used with any
image that you want to test. There is nothing special about the GLM in
SPM. All SPM does is to run a GLM (same as a stats program), except it
does it 500,000+ times - once at each voxel. Because of matrix
algebra, SPM does one slice at a time, rather than one voxel. In the
future, I suspect it will do the whole brain at one time and that will
speed it up.
Hope this helps.
>
> Thank you very much,
>
> Janosch
|