Dear All,
I have recently encountered an issue that I wanted to discuss.
Like others, we have started doing our statistical analyses in
native-space and then only normalize the con-images, using the
deformation field from, e.g., DARTEL or unified segmentation. We then
perform the RFX group analyses in normalized space, which should be
equivalent to normalizing the individual functional images (let's ignore
differences due to interpolation or definition of smoothing widths in
native vs. normalized space).
I have now come across an analysis where the resulting second level mask
included much of the volume, and when checking the con-images, this was
due to the value of the normalized non-brain voxels not quite being
zero, but only very close to it (as the original value of NaN in the
native space con-image is interpolated during spatial transformation). I
must admit that, personally, I found it counterintuitive that native
space values of both 0 and NaN are interpolated to be close to 0 in the
normalized images, when using spm_defs. This effectively increases the
search space for multiple comparisons on the second level, which
substantially influenced the smoothness estimation, as naturally, these
non-brain regions were very smooth.
There are of course two obvious solutions: one (easier), use an explicit
mask on the second level. Two, somehow keep the spatial transformation
algorithm from interpolating all voxel values it finds in the native
space maps. I looked into what happens in spm_defs in the apply_defs
function but could not easily figure out how to achieve this, and was
afraid I'd break something. Interestingly (to me), it looks like the
problem is aggravated when using higher order bsplines interpolation, so
setting the interpolation for spm_bsplinc to 0 would be another way to
circumvent this, but I don't think this is the best solution.
The bottom line of this long mail therefore is, does anyone have a good
solution for either keeping the deformation utility from interpolating,
say, NaN, or ideas for another good workaround? As always, your esteemed
input is much appreciated :)
Cheers,
Marko
--
____________________________________________________
PD Dr. med. Marko Wilke
Facharzt für Kinder- und Jugendmedizin
Leiter, Experimentelle Pädiatrische Neurobildgebung
Universitäts-Kinderklinik
Abt. III (Neuropädiatrie)
Marko Wilke, MD, PhD
Pediatrician
Head, Experimental Pediatric Neuroimaging
University Children's Hospital
Dept. III (Pediatric Neurology)
Hoppe-Seyler-Str. 1
D - 72076 Tübingen, Germany
Tel. +49 7071 29-83416
Fax +49 7071 29-5473
[log in to unmask]
http://www.medizin.uni-tuebingen.de/kinder/epn/
____________________________________________________
|