Kalina Christoff wins (or maybe I just misunderstood you, Darren)! She
wrote the following:
>Did you try entering "none" at the Threshold masking prompt,
>"no" at the Explicit masking prompt, and "omit" at the Global
>calculation prompt? It seems to me one or more of these options, when
>specified differently, can cause an empty mask to be produced for a
>2-sample test at the group level of analysis.
This did the trick. I was thinking only about explicit/implicit masks, and
I had been putting in a threshold mask of .8 (but on reflection, I guess
that the con.img's must have this threshold already applied to them when
written by the first level analysis).
Thanks both for lending part of your Saturdays!
-David
-----Original Message-----
From: Darren R. Gitelman [mailto:[log in to unmask]]
Sent: Saturday, September 30, 2000 4:17 PM
To: Kareken, David A.
Cc: [log in to unmask]
Subject: RE: 2-sample t-test
David:
Hmmm. You may have to wait for an answer from an actual expert. Make sure
you actually look at all the con**.img files to make sure they all contain
data though it sounds like you've done that. Just to make sure all the
origins are correct you might try looking at the con**.img files (did they
come from different analyses?) using the check Reg function. You can look
at up to 15 images this way and it will be instantly apparent if one
doesn't line up.
Did you set up the analysis through the basic models choice or through PET
statistics? I've found that basic models choices are in general configured
properly to deal with the format of the con**.img files. These files have
both pos. and neg values and outside the area of brain data the images
contain NaN values (not a number). With the PET statistics choices you have
to threshold the images at -Inf I believe to get the mask to work properly.
Hope this really helps,
Darren
>Thanks for the help, Darren. Unfortunately, changing the F threshold did
>not change anything. I see no reason why the origins of the images should
>be a problem, and they appear to give perfectly reasonable (anatomically
>correct) answers when treated as one entire group. I also avoided an
>explicit mask in all the analyses.
>
>Any other ideas?
>
>-David
>
>-----Original Message-----
>From: Darren R. Gitelman [mailto:[log in to unmask]]
>Sent: Saturday, September 30, 2000 3:04 PM
>To: Kareken, David A.
>Cc: [log in to unmask]
>Subject: Re: 2-sample t-test
>
>
>Dear David:
>
>I think the message means that there ended up being no voxels whose
>significance passed the Upper tail F probability threshold. (Assuming that
>all con**.img images are appropriately oriented with the proper origins,
>etc. Also make sure you haven't masked the images at the second stage of
>the analysis. I suggest going into preferences and changing your F
>threshold to 1 (= no threshold) and rerunning your stats.
>
>HTH,
>Darren
>
>
>
>>SPMers,
>> Can someone help with the following message? I have a two group PET
>>activation study. I used a Condition x Subject model to generate
con.img's
>>for each subject to take to a second (RFX) level analysis. When I enter
>>these two groups into this model, however, the analysis proceeds until the
>>last image plane/plank, and produces the following error output.
>>
>> Warning: No inmask voxels - empty analysis!
>> > In C:\Program Files\spm99\spm_spm.m at line 1074
>> In C:\Program Files\spm99\spm_spm_ui.m at line 1427
>> Design parameters : ...intrinsic
>>autocorrelation ...V, & traces ...scaling
DesMtx
>>...closing image files ...smoothnessWarning: Divide by
>zero.
>> > In C:\Program Files\spm99\spm_spm.m at line 1137
>> In C:\Program Files\spm99\spm_spm_ui.m at line 1427
>> Warning: Divide by zero.
>> > In C:\Program Files\spm99\spm_spm.m at line 1138
>> In C:\Program Files\spm99\spm_spm_ui.m at line 1427
>> ??? Index exceeds matrix dimensions.
>>
>> Error in ==> C:\Program Files\spm99\spm_spm.m
>> On line 1142 ==> FWHM = FWHM*((RESEL/prod(FWHM(1:N))).^(1/N));
>>
>> Error in ==> C:\Program Files\spm99\spm_spm_ui.m
>> On line 1427 ==> spm_spm(VY,xX,xM,F_iX0,xC,xsDes)
>>
>> By the first line, it appears to suggest that there are no voxels to
>>analyze. I submitted each of the groups separately, as well as one entire
>>group, to a one sample t-test to test the con.img's against a null
>>hypothesis and it works fine, producing the expected results. Thus, there
>>doesn't appear to be a particular problem with the images themselves.
>>
>> Suggestions?
>>
>>Thanks all.
>>
>>________________________________
>>David A. Kareken, Ph.D., ABPP
>>Assistant Professor & Director
>>Neuropsychology Section, Department of Neurology
>>RI 5999C
>>Indiana University School of Medicine
>>Indianapolis, IN 46202
>>(317) 274-7327 (voice)
>>(317) 274-1337 (fax)
>
>-------------------------------------------------------------------------
>Darren R. Gitelman, M.D.
>Cognitive Neurology and Alzheimer's Disease Center
>E-mail: [log in to unmask] WWW:
>http://www.brain.northwestern.edu
>Voice: (312) 908-9023 Fax: (312) 908-8789
>Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
>-------------------------------------------------------------------------
-------------------------------------------------------------------------
Darren R. Gitelman, M.D.
Cognitive Neurology and Alzheimer's Disease Center
E-mail: [log in to unmask] WWW:
http://www.brain.northwestern.edu
Voice: (312) 908-9023 Fax: (312) 908-8789
Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
-------------------------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|