Hi Martin,
I suspect this happening because SPM sets the threshold for analysis
to be at 80% of the global mean.
In spm_fmri_spm_ui.m line(410)
try
TH = g.*gSF*defaults.mask.thresh;
catch
TH = g.*gSF*0.8;
end
Unless you apply session specific scaling, gSF = GM./g and GM =100; So
the threshold is always at 80.
Load SPM.mat for a subject and check what the value of SPM.xM.TH is. I
suspect this is 80 (unless you apply session specific scaling). Make
sure this is -inf, save the SPM.mat file and rerun the estimation
step. The primary reason for the scaling I believe is to keep out
outliers. When the ROI mask is small, and anatomically defined, such a
rigid threshold may not be necessary.
Hope that helps.
Satra
--
Satrajit Ghosh, PhD
Speech Communications Group
Research Lab of Electronics, MIT
On 2/5/06, Martin Zalesak <[log in to unmask]> wrote:
> Dear all,
>
> I am sorry to be sending this again, but I really want to find out
> the answer to this problem. I think this could be addressed by
> someone who has detailed knowledge of the SPM2 machinery.
>
> I am trying to run the SPM GLM on native space BOLD images as
> follows: I created an anatomical ROI on native space anatomical image
> first. Then I applied this ROI as a mask to all my BOLD images, thus
> creating partial brain BOLD volumes in native (i.e. non-normalized) space.
>
> Now I want to run the GLM on these partial brain images to obtain the
> beta parameter estimates. When I check the results, I see that the
> resultant analysis mask (mask.img) looks very much like the
> anatomical mask that I used in the first place, which is good.
> However, when I look at the beta images, I notice that most of the
> voxels within the mask (defined by mask.img) have NaN values. I
> thought that all voxels within the space defined by the mask should
> have non-zero beta values.
>
> That's a problem as I need to get a beta value for every voxel that
> is inside the mask. Does anyone have any ideas or suggestions? How do
> I fill the whole ROI with beta values? Does this have anything to do
> with the fact that I am not using smoothing and normalization?
>
> Please advise. Thank you,
> Martin
>
|