OK.  By "baseline shift" I was thinking of between-session scanner drift.

You're right, if the actual brain state is different between e.g. fixation in different sessions, then a lot of conventional wisdom doesn't apply.

Stephen J. Fromm, PhD
Contractor, NIMH/MAP
(301) 451--9265
________________________________________
From: MCLAREN, Donald [[log in to unmask]]
Sent: Wednesday, November 04, 2009 3:27 PM
To: Fromm, Stephen (NIH/NIMH) [C]
Cc: [log in to unmask]
Subject: Re: two sessions analysis

On Wed, Nov 4, 2009 at 1:41 PM, Stephen J. Fromm <[log in to unmask]> wrote:
> On Tue, 3 Nov 2009 16:37:26 -0600, MCLAREN, Donald
> <[log in to unmask]> wrote:
>
>>I'd look at the beta values for both tasks in these regions that you
>>expect differences.
>>
>>If you set up an F contrast:
>>1 0 0 0
>>0 1 0 0
>>
>>This will allow you to plot values of both tasks to see how close the
>>beta values are. In your comparison between conditions, you are
>>comparing the beta values -- not the T-statistics which you see. For
>>example, its possible that there is a lot of subthreshold activity
>>bilaterally.

column1: task 1 run 1
column2: task 2 run 2
column3: baseline run 1
column4: baseline run 2

column 1 and column 2 are both with respect to the basline term and
take into account the session effects -- however, it doesn't control
for a baseline shift (see below).


>>
>>As for combining them in a single run -- this would help because you
>>don't have to worry about the baseline being different between runs.
>>In your case, you can't be sure that differences are due to your task
>>or due to a baseline shift.

I meant from an experimental point of view, not an analysis point of
view. One should not combine runs into a single run when doing the
analyses. A good example of why combining tasks into a single run
during acquisition -- not analysis -- is as follows.
Three runs were performed with the identical stimuli, except that the
subject was instructed to focus their attention on narrowly or broadly
or not at all during the fixation periods. The fixation periods were
then analyzed and the activity was found to be different in each run.
Now, if we flip that around and look at the activation of the task in
these same regions, we would conclude that the tasks between runs were
different (tasks were the same, baseline was different). Now this was
an experimental manipulation, but one could easily imagine that the
baseline (fixation/blank screen) could be different between runs
depending on the task. In fMRI, everything is relative, so you don't
know if the baseline or task is causing the differnce.



>
> In generally, combining into one run isn't a good idea, unless it's done very
> carefully, because of technical issues like high-pass filtering, the modeling of
> the hemodynamic response, etc.

Auto-correlation as well.

>
> Furthermore, contrasts that take differences _between_ sessions that don't
> account for the main effect of session (and hence don't control for baseline
> shift) shouldn't be performed anyway.

I totally agree.


>
> Best,
>
> S
>
>>Best Regards, Donald McLaren
>>=================
>>D.G. McLaren
>>University of Wisconsin - Madison
>>Neuroscience Training Program
>>Office: (608) 520-0586
>>=====================
>>This e-mail contains CONFIDENTIAL INFORMATION which may contain
>>PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED
>>and which is intended only for the use of the individual or entity
>>named above. If the reader of the e-mail is not the intended recipient
>>or the employee or agent responsible for delivering it to the intended
>>recipient, you are hereby notified that you are in possession of
>>confidential and privileged information. Any unauthorized use,
>>disclosure, copying or the taking of any action in reliance on the
>>contents of this information is strictly prohibited and may be
>>unlawful. If you have received this e-mail unintentionally, please
>>immediately notify the sender via telephone at (608) 520-0586 or
>>email.
>>
>>
>>
>>On Tue, Nov 3, 2009 at 4:30 PM, Stephen J. Fromm <[log in to unmask]>
> wrote:
>>> On Tue, 3 Nov 2009 12:22:12 +0000, Jonathan Peelle <[log in to unmask]>
>>> wrote:
>>>
>>> Jonathan,
>>>
>>> Maybe I'm interpreting Marta's design matrix schematic incorrectly, but
> doesn't
>>> it look like there's an implicit baseline in each of the sessions?  (The dark
>>> bands in the regressors for the conditions.)  If so, that would serve as
>>> your "C" below.
>>>
>>> Of course, everything you wrote stands on its own.
>>>
>>>>Hi Marta
>>>>
>>>>The problem is that the effect in which you are interested (right vs.
>>>>left ankle dorsiflexion) is also a difference between sessions.  By
>>>>modeling session effects (which is the right thing to do) (columns 3
>>>>and 4 of your design matrix), you essentially remove any overall
>>>>difference between sessions, making it difficult or impossible to pull
>>>>out differences in your conditions.  Put another way, your effects of
>>>>interest are confounded with session effects.
>>>>
>>>>If you have a chance to change the design, you could consider one of
>>>>the following alternatives:
>>>>
>>>>1) Have both conditions of interest in both sessions (i.e. alternate
>>>>left and right ankle movements within both sessions).  This way you
>>>>have the same number of events of each type but they are not related
>>>>to session effects.
>>>>
>>>>2) Have some baseline condition you can compare the ankle dorsiflexion
>>>>to.  This is no doubt explained in more detail somewhere previously on
>>>>the list, but the idea is that it's possible to look at an interaction
>>>>across sessions, but not really main effects.  I.e.
>>>>
>>>>Session 1: condition A and condition C
>>>>Session 2: condition B and condition C
>>>>
>>>>contrast A > B is problematic because of session effects;
>>>>contrast: (A > C) > (B > C) would be fine.
>>>>
>>>>If you are stuck with the data as it stands (and no possibility of
>>>>finding a baseline condition to add to the model), I don't know if
>>>>there is a particularly good solution.  You can choose not to model
>>>>session effects, but this will add noise, and I think be a bit harder
>>>>to interpret (e.g., is higher signal in a region actually due to your
>>>>task, or could it just be a byproduct of one session by chance having
>>>>a different level of activity than another?).
>>>>
>>>>Hope this helps,
>>>>Jonathan
>>>>
>>>>
>>>>On Tue, Nov 3, 2009 at 11:46 AM, Gandolla Marta
>>>><[log in to unmask]> wrote:
>>>>> Hi everyone,
>>>>>
>>>>>    I have some problems in two sessions analysis. I want to compare two
>>>>> activation maps from the same subject in two different conditions. I built
>>>>> the design matrix with two sessions (using first level analysis) so I ended
>>>>> up with four columns as you can see from the first figure of the attached
>>>>> file. I then did inference analysis with contrast vector of [1 -1 0 0] and I
>>>>> suppose I shoud get a map with the significative differences between
> the
>>> two
>>>>> conditions.
>>>>> my problem is that if I implement this same approach with maps that are
>>>>> significantly different for sure (right ankle dorsiflexion and left ankle
>>>>> dorsiflexion) I get a "difference map" that is not at all as expected. so as
>>>>> you can see what I'm talking about, the second figure of the attached
> file
>>>>> is the result I got.
>>>
>
>
>