>Date: Thu, 10 Feb 2000 14:45:23 -0500
>To:Kewei Chen <[log in to unmask]>
>From:[log in to unmask] (Badreddine Bencherif)
>Subject:Re: spm-processed data behavior of longitudinal PET study
>
>Dear Kewei:
>>As a post-SPM analysis, we extracted the voxel values (spm-normalized
>>and corrected) for the
>>two scans of each subject from
>>various locations. (i.e, XA(:,i) for XYZ(:,i) location)
>>For each given location, we have a pair of pixel values for each subject:
>>p1 for baseline scan and p2 for the follow-up scan.
>>A scattergraph was made by using matlab commands:
>>What we found is that,for example with group B,
>> for all the locations that we have examined so far,
>>highest p1 is always paired with the lowest p2, and
>>lowest p1 is always paired with the highest p2.
>>The pairs in the middle follow a similiar trends.
>>The attached tif file is such an example.
>
>
>I think it is due to the assumption of equal distributed error implicit in
>SPM. If you compute the standard deviation of each condition p1 and
>condition p2 at a given voxel location they will be absolutely identical.
>
>I found similar results, highest bas always paired with lowest act and
>lowest bas always paired with highest act and identical standard
>deviations.
>SEE BELOW
>I recently posted a similar question
> at voxel -56, 10, 28 for
> bas 128.87 +/- 2.4684 (mean +/- std)
> act 114.56 +/- 2.4684
> and for another voxel at -52, 10, 26
> bas 161.09 +/- 4.9552
>act 149.65 +/- 4.9552.
>Interestingly plot of act along the y axis ( . . . . . . ) is
> identical but inverted compared to bas ( . . . . . . ) and shifted
>down. Obviously variance of bas versus act is the same for same voxel location
>and is heterogeneous between voxels.Can someone explain why the variance
>between bas and act for a given voxel location is absolutely identical ?
>Is it a constraint inherent to the math of SPM or possibly a simple
>statistical relationship that I disregarded?
>
>Answer from Dr Zarahn was
>It is actually a combination of both. The constraint (or more correctly, the
>assumption) inherent in the modeling implemented by SPM is that of
>identically distributed errors. This implies that the time series errors
>are the same during each condition. In addition to the properties of the
>time series errors, the variance of a condition effect also depends on
>the design matrix. It can be induced that there were an equal number
>of observations for bas and act, as in that case the design matrix
>factor would be equal for bas and act. If, for argument's sake, the number
>of observations associated with bas was greater than that associated
>with act, then you would see that the variance of the bas parameter estimate
>was smaller than that associated with act.
>
>
>Didine
>
>
>
Badreddine Bencherif, MD
Division of Nuclear Medicine
Johns Hopkins University School of Medicine
601 N. Caroline St. / JHOC 4230
Baltimore, MD 21287-0855
Phone: (410) 614-2787
Pager in hospital: 3-2050 out hospital: (410) 283-2050
Fax: (410) 614-1977
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|