Dear List: I posed a question earlier this month about plotting mean signal
values from a cluster of voxels in a PET analysis (SPM99). I want to use these
mean signal values in SEM to assess connectivity. In brief, I used marsbar (v
0.23) to define a cluster of activated voxels from my group analysis (10
subjects) and saved it (marsbar saves the cluster as a ROI matlab file). I
then used marsbar to plot the mean signal from that cluster in each of my 10
individual subjects. However, it is possible (even likely) that some or all of
the voxels in that cluster were not significant in individual subjects' data.
David Corina had written the group near that time about the fact that SPM99
does not plot mean signal values from non-significant clusters, so I was
concerned that the mean signal values I was obtaining from marsbar were not
from the cluster I specified. Dr. Matthew Brett responded to my original
question with: "MarsBaR uses all the voxels in your cluster, regardless of
whether they were significant in the individual subject." In contrast, David
Corina responed: "What we have seen is that despite setting a specific ROI in
Marsbar, if this falls within a non-signifincant area the time course returned
is from the nearest significant cluster (a threshold previously defined in your
>model). This is the same behavior seen in SPM (as Marsbard apprears to
>simply evoke the plotting function from SPM). WHat is confusing and
misleading is that graphing the marsbar result will indicate on the Y axix that
the plot is from the ROI you specifed rather than the "jumped to" significant
cluster." A colleague and I have since attempted to directly obtain (in
matlab) the mean signal in a matlab cluster file generated by marsbar in a
single subject's data . However, the mean signal values obtained this way are
several orders of magnitude above those returned by marsbar, presumably due to
the global normalization and proportional scaling applied in the SPM99
analysis). Is there a way to test whether I am obtaining the correct mean
signal values in the single subject marsbar analysis when I have defined a
significant cluster from the group analysis?
Thanks,
Warren Darling
|