[Marsbar-users] Re: Marsbars

[Matthew B. <ma...@mr...>]

Hi,
 
> I did some typical SPM omnibus analyses using snr files and chose ROIs based
> on those analyses.
> 
> 1) I assume I choose the snr files to place the ROIs on?

Well, that's an interesting question.  It depends a bit on the ROI.  If you 
have a very precisely defined ROI - for example for the basal ganglia, from an 
anatomical scan - then you might want to use the not-smoothed data - in this 
case the nr images.  This is the function of "convert to unsmoothed" in the 
statistics menu.  It takes an SPM design that was run on smoothed data - here 
snr images - and makes a new analysis definition file (e.g. SPMcfg.mat) that 
points at the not-smoothed versions of the same files - here nr images.
Of course, for this to work, you need to have the nr images in the same 
directory as the original snr images.

But usually the ROI is not very precisely defined - for example from 
activation data after heavy smoothing - and using the snr files for the ROI 
analysis is fine.

> 2) What is the difference between scaled and raw? I'm guessing that scaling
> is at least roughly quivalent to proportional scaling in omnibus analyses?

Scaling involves several possible options; each uses the thresholded mean 
voxel value (TMVV) - see http://www.mrc-cbu.cam.ac.uk/Imaging/spm_stats.html

Session specific scaling: the mean TMVV is scaled to be the same between 
sessions.  The mean TMVV will be set to equal the grand mean (see below).
Proportional scaling: the TMVV for each scan is made the same by dividing the 
TMVV into the whole image.  And then multiplied by the grand mean (see below)
Raw: no adjustment across scan and session.

> If you choose raw is does this provide true absolute instead of normalized
> values for the territory in the ROI?

Well, yes, true absolute in the sense that the values are in the units of the 
images themselves, without further scaling, if you choose 0 for the scaling 
parameter below.

> 3) Do you have to choose 0 for raw? If you choose, say 50, is this a
> separate method of normalization?

Choosing 0 gives you the values in the same units as are in the images.  
Choosing 50 will scale the average of the globals for all images to be 50 - so 
that gray matter signal will usually be at about 100 for example.

> 4) And lastly, what do the values provided by data extraction represent? I
> have values ranging from .7 to 2.3. Is 1.0 equivalent to the global mean (if
> normalized)? And what do the values represent if raw?

Well, that will depend on what scaling you chose, but get back to me if this 
is not clear,

Best,

Matthew