Just a little extra comment to this thread,
> Under ideal conditions, where the PVE was very accurate
> and there was little spatial blurring in the original image (which
> generally is impossible due to the finite Fourier samples taken,
> as well as T2/T2* blurring) then the "right" thing to do would
> be to do the most accurate interpolation for the rotations, etc.,
> followed by an integration over the new voxel area, as I
> suggested. However, things aren't ideal, obviously, so the
> details here make less difference.
In the next patch of FSL we'll have an updated version of applywarp
that makes it very easy to do precisely what MJ describes above. The
command
applywarp --ref=my_ref --in=my_scan --out=scan_in_space_of_ref --
premat=my_flirt_matrix.mat --super --interp=spline
will take "my_scan", transform the coordinates using
"my_flirt_matrix.mat", interpolate using cubic B-splines to a matrix
with twice the resolution of "my_ref" and then integrate up over 8
neighboring voxels to produce "scan_in_space_of_ref".
The supersampling+integration is performed when the --super flag is
set, and is useful precisely when you want to resample a high-res
image into the space of something with lower resolution. If the
resolution of the images is _very_ different you can complement it
with --superlevel=4 which will use an intermediate image with 4 times
the resolution of --ref before integrating over the neighboring 64
voxels.
The --interp=spline flag is independent of if you want to use
supersampling or not and simply specify the use of cubic B-spline
interpolation. So no need to switch to SPM anymore ;-)
Jesper
|