Tobias
> I have a problem with activation clusters extending outside the brain.
>
> I performed a second-level Rfx-analysis according to the instructions in the
> SPM Help-facility and many helpful comments in the discussion list archives.
>
> Things worked really well but the results were rather thin. After
> reading that greater smoothing could enhance the results, I applied
> additional smoothing to my first-level con*-files, as recommended in
> a correspondance in the archives.
>
> Indeed, the results got better, i.e. the cluster sizes and significances
> increased. Unfortunately, the clusters are now extending
> considerably outside the brain.
>
> My understanding was, that this should not happen after normalizing
> my images to the same template before first-level analysis.
>
> Now, how is this possible ?
By smoothing, you've "created" data outside of the brain; the new data
are a mixture of zeros (from outside the brain) and nonzero data from
within the brain
> What can I do to get rid of such "spurious" activated voxels outside
> the brain ?
Create a mask based on each of the subject's brain masks ("mask.img").
This is important to do, not only to eliminate voxels outside the
brain but to increase your power.
You gain power because your corrected significances reflect the extra
search over the "cloud" of these new extracerebral voxels; when you
eliminate this cloud via masking, you will have fewer resels/voxels to
search over, and hence a less severe multiple comparisons correction.
Below is a code snippet which will create an intersection mask based
on each of your intrasubject masks (assuming you have run intrasubject
analyses on the exact same data---that is, you spatially normalized
your raw data.)
-Tom
-- Thomas Nichols -------------------- Department of
Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
%
% Create an intersection mask
%
% Doesn't check that all images have same dimensions or have the same
% space.
%
fNm = 'mask_all.img'
P = spm_get(Inf,'mask.img');
n = size(P,1);
str = sprintf('i%d&',1:n);
str(end) = [];
spm_imcalc_ui(P,fNm,str,{[],[],spm_type('uint8')});
|