Dear all,
I am still trying to figure out why the use of a mask is not
permitting the estimation of this particular analysis.
From spm_spm.m these lines seem to be important:
------------------------------
if ~isfield(xVi,'V')
%-F-threshold & accumulate spatially whitened Y*Y'
%----------------------------------------------------------
j = sum((Hsqr*beta).^2,1)/trMV > UF*ResSS/trRV;
j = find(j);
if length(j)
q = size(j,2);
s = s + q;
q = spdiags(sqrt(trRV./ResSS(j)'),0,q,q);
Y = Y(:,j)*q;
Cy = Cy + Y*Y';
end
end % (xVi,'V')
------------------------------
The 's' variable later checked for significant voxels.
As I said before, the same mask works when I split the analysis in two
2x1 anova-s. Or simply if I remove the 'subject' factor from the
design to leave only four columns (2x2).
I am attaching the display of the *mask* and *con* images. Do you see
anything wrong? Somebody knows how can I overcome this problem?
Thank you and good Friday.
Dorian
2009/10/20 Dorian P. <[log in to unmask]>:
> Hi,
>
> The mask works for other analysis. I created it with Imcalc (i1>0).
> However, there are non-1 values especially at the border (even after
> running Imcalc again).
>
> Tried also setting "Implicit Maks: No". Yet, it didn't help.
>
> Strangely, if I split the 2x2 design in two 2x1 designs the analyses are OK.
>
> Any other suggestion?
> I will try to debug as Darren suggested (hoping I don't get lost in
> SPM variables).
>
> Dorian.
>
>
>
>
> 2009/10/20 Mohammed Al-Rawi <[log in to unmask]>:
>> Hi
>> May be you can try a mask with all 1's....and see if the problem persists
>>
>> Regards
>> Al-Rawi
>>
>>
>> ________________________________
>> From: Darren Gitelman <[log in to unmask]>
>> To: [log in to unmask]
>> Sent: Tue, October 20, 2009 4:37:44 PM
>> Subject: Re: [SPM] No significant voxels because of the mask
>>
>> Dorian
>>
>> Are you sure your explicit mask lines up with your images? Are the
>> values in your mask >0?
>>
>> I've checked the code and there doesn't seem to be an interaction
>> between the various masks, e.g., an explicit mask shouldn't raise the
>> threshold for implicit masking and vice versa. All the masks are
>> intersected to give a final mask. Still, if you have other types of
>> masking enabled you might try turning them off but leaving on the
>> explicit mask and re-running the analysis.
>>
>> If you want to try to debug this, if no one else has other ideas, then
>> you would look at line 680 in spm_spm and check if Cm is ever
>> non-empty (these are the values loaded from your explicit mask). Then
>> line 702 in spm_spm sets Y to the in-mask data (before any stats are
>> done, but after all masking and thresholding).
>>
>> Darren
>>
>>
>> On Mon, Oct 19, 2009 at 8:14 PM, Dorian P. <[log in to unmask]> wrote:
>>> Dear all,
>>>
>>> I am running a flexible factorial with an explicit mask. The
>>> estimations stops with error:
>>>
>>> -------------------------------
>>> Error running job: Error using ==> spm_spm at 861
>>> Please check your data: There are no significant voxels.
>>> ------------------------------
>>>
>>> Nevertheless, when I remove the explicit mask, the estimation runs and
>>> there are significant results (p < 0.001) in the areas that the mask
>>> covers.
>>>
>>> Why does this happen? I thought using a mask the analysis would be
>>> more sensible, not less.
>>>
>>> (using the last revision SPM5 r.3381)
>>>
>>> Thank you.
>>> Dorian
>>>
>>
>>
>
|