Print

Print
Dear Carles,

This is likely caused by some sort of outlier.  The lucky thing is that you have a good candidate for which voxel is the problem... i.e. the voxel with a T-value over 1000.  If I had to guess I would say it is an extreme voxel right at the boundary with white matter.

To find the problem, I'd make two images, Stdev (ImCalc: Input ResMS.img, Expression: sqrt(i1)) and FWHM (ImCalc: Input RPV.img, Expression: i1^(-1/3)).   The standard deviation image might show various problems, but if nothing else you can check that problem location (30  −9  18) and see what's going on there.  Also check the mask (clearly visible in either image), to ensure you're not counting lots of non-brain voxels (which will also screw up your resel estimate).  The FWHM tells you the local average FWHM in voxels, and may also suggest some problems (usually non-brain or non-GM voxels contributing to FWHM causes trouble).

Once you've found problem voxels, I'm afraid there's no easy way except to check all the subjects at that location and see who's causing the trouble.  The fix may be to drop subjects or, via an explicit mask, drop voxels.

All of these steps are actually a subset of what my SPMd toolbox does (but with plotting tools so you can easily see what subject is causing the trouble), if you want to give it a try.

-Tom


On Sat, Jun 21, 2008 at 1:24 AM, CARLES M. FALCON FALCON <[log in to unmask]> wrote:
Dear spmers,
The statistical parametric map I've got from a longitudinal DARTEL-VBM study
presents a singular point with huge T-value (see attached file). The expected
resel size is 0 so the number of expected resels is also huge. The estimated
smoothness is zero.
I followed DARTEL protocol as set in the manual except for normalization to MNI
template step, that I used both affine and not affine transformations to get a
better matching between templates. I run SPM5 (updates_1782) on Matlab 7.5.0

Any clue about how to solve this problem will be much appreciated.

Yours

Carles Falcon
Unitat d'Anàlisi per la Imatge-IDIBAPS. Hospital Clinic Barcelona
Unitat de Biofísica i Bioenginyeria. Universitat de Barcelona
Casanova 143. 08036 Barcelona. Spain


-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/




--
____________________________________________
Thomas Nichols, PhD
Director, Modelling & Genetics
GlaxoSmithKline Clinical Imaging Centre

Senior Research Fellow
Oxford University FMRIB Centre