Dear Grant,
>Thank you for your prompt and detailed reply. I am glad to hear that I
was not
>too far off base.
>But your answer raised another question. How does SPM handle missing values
>due to pixels falling in and out of the grey matter threshold on different
>scans/subjects
>
>Lets say that for pixel Xij (i = subject and j = scan) and a grey matter
>threshold of 0.8 we have the following distribution of values across
subjects,
>with pixel values expressed as fraction of Global Mean
>
>
>Subj Scan 1 Scan 2
>1 1.2 1.4
>2 0.75 0.9
>3 1.1 1.3
>4 0.74 0.78
>
>How does SPM treat this data ? Subj 2, scan 1 falls below the grey matter
>threshold and therefore would be excluded from the search volume. On
Subject 4,
>the pixel falls below the grey matter threshold on both scans. I can think of
>several alternatives.
>
>A) Subject 2 and 4 are eliminated from the analysis entirely, i.e. all
voxels
>are dropped for both of their scans
>B) This specific pixel is dropped, leaving 2 subjects for this pixel Xi1,
but 3
>subjects for pixel Xi2 and all other pixels.
>C) This specific pixel is given a value of 0 for Scan 1 and 0.9 for Scan
2 for
>Subject 2, and values of 0 and 0 for Subject 4.
>D) This specific pixel is given a value of 0.8 (minimum grey matter) for
Scan
>1 and 0.9 for Scan 2 of Subject 2 and 0.8 and 0.8 for Subject 4
>
>None of these seem exactly right, but B) seems the least problematic
except for
>having different degrees of freedom for different pixels.
>
None of the above is actually correct. For any given voxel it is required
that the condition described in my previous reply is fulfilled, if not this
VOXEL is dropped from the analysis completely. In the best of worlds (one
in which spatial normalisation is perfect) this would obviously not be a
problem. In reality though there will be a "gray zone" where a given voxel
will be in gray matter for one subject and outside the brain for another.
>From a specificity perspective (avoiding false positives) it is obviuosly
then better to exclude this voxel from the analysis completely. Also, as
anyone who has worked with PET and pain will tell you, there are certain
conditions in which there will be highly task correlated changes in
extracerebral (scalp) perfusion. Using a too liberal criterion for defining
"intracerebrality", these areas would then show up as highly significant
"activations" and may be interpreted as being in the brain. If you think
such a mistake would be easily spotted, I might mention that very
experienced workers have on occasion attributed changes in jaw muscle
perfusion to funcional changes in temporal cortex resulting from anxiety.
>This is not a theoretical issues. I have noted such patterns with FDG
scans in
>areas such as the medial temporal lobe (amygdala).
>
I would love to hear specifically what those patterns are.
Good luck Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|