I think that it's clear that the problem is with proportional scaling, when your
intervention has such a profound effect on global cerebral blood flow. If you
are using quantitative rCBF images, as you appear to be doing for the "separate
ROI analysis" then you should not perform any global normalization. You can
only get away with global normalization when the global cerebral blood flow is
not changing appreciably due to changing conditions.
Kaisti Kaike wrote:
> Dear SPM'ers
>
> I would be most grateful for any advice or comments. (Sorry to repeat
> myself,
> but my previous mail (29th Sept.) didn't seem to elict any responses.)
>
> We're using [15O]H2O- PET to study effects of deep anesthesia on rCBF.
>
> The problem is that the results of SPM subtraction analysis (anesthesia vs.
> awake) show large clusters of "relatively increased flow" situated in white
> matter. Clusters are partly masked out by Analysis treshold (0.8).
>
> Quite frankly, I'm not sure how I should feel about such results. In certain
> sense they are logical as anesthesia does reduce flow more in gray than in
> white matter. [A separate ROI analysis on quantitative images showed that
> regional flow at awake was roughly 65 (ml/100g/min) in gray matter and 25 in
> white matter. During anesthesia flow values were reduced to 29 and 14,
> respectively. Thus, the gray/white-matter flow ratio reduced from 2.6 to 1.8
> (30% reduction).]
>
> My main concern is that this kind of major "shift" might affect the validity
> of global flow scaling and therefore create a see-saw effect that
> compromises all the statistics. Any comments? We used mean voxel value (w.
> fullmean/8 mask) for global flow calculation, and proportional scaling for
> global normalisation.
>
> I couldn't find anything quite similar from the e-mail archive. I would
> highly appreciate any suggestions - or questions.
>
> Yours,
> Kaike
|