>Dear Dr Anderson:
>
>I would like to thank you first for your comment and explanations.
>I would like to discuss more about the following:
>A)
>>In addition, since you will (through your modelling) have removed the
>>obviously confounding sources of global variation, it may actually be that
>>you should not do any global normalisation at all. This is especially true
>>if your "activation" is of a more profund type, such as a pharmacological
>>intervention.
>
>I would like to point out that after doing our modelling the range of
>globals is still quite large (0.70 to 1.40) with a good intrasubject
>reproducibility if subject 1 for example has a mean global baseline of let
>see 0.9 his activation will be close to 0.9 and for subject 2 if baseline
>mean global is 1.20 then activation is close to 1.20. In addition the
>(pharmacological or other ) intervention that we did not shift the range of
>globals. In other words the range of globals was for the 10 baseline
>(0.54-1.51) and activation (0.62-1.33 which is a subset of the baseline
>range of globals 0.54-1.51).
>Do you still believe that a normalization for the globals should not be
done ?
>
Generally speaking the results you obtain from normalised data vs.
unnormalised data will answer different questions. "Activations" in
normalised data are areas that change its "activity" in response to
stimulation in a manner that is different from that of "most" areas of the
brain. "Activations in unnormalised data are areas that change its
"activity" in response to stimulation, period.

In a sense I guess it is the latter question that most of us want to
answer, and it would therefore seem logical not to perform any
normalisation. However, this means that we must in some other way account
for changes in injected activity and "true" global flow, and that typically
involves arterial cannulation. For that reason most people use
normalisation to answer the first question above, and by assuming that
"true" global flow does not change as a consequence of stimulation and that
the local activations does not appreciably change our estimate of global
flow, they will interpret their results as if they were answering the
second question. 

For a majority of all studies this is a perfectly feasible assumption and
the strategy works better than to actually try to correct for "globals" by
modelling the flow (due to inevitable statistical uncertainty in the
modelled results).

I cannot really guess which will be the best in your case, but given that
you have modelled your data I certainly think you should give both options
a try. The fact that you have good intrasubject reproducibility indicates
that you may have a good sensitivity even without normalisation (provided
you use a paired statistical test).

>B)
>It is slightly surprising that you
>>found a positive correlation between slope and intercept (I would have
>>expected a negative) but I guess the explanation for that might be that
>>they both covary positively with the pixel values.
>
>I had the same reaction as yours and I go along the same explanation that
both
>covary positively with the pixel values.
>
Fair enough.

				Good luck Jesper



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%