Dear Klaas,
thanks for your answer! Unfortunately, I obviously didn't make myself
clear enough: The problem I am dealing with is the simple fact that some
subjects do not show the effect within a given radius from the maximum
as identified in the RFX-analysis, even when using a liberal threshold
of .05 uncorrected. I am quite confident that this is a rather common
finding, especially when one studies "higher-level" cognitive processes.
So my question simply is: how do people deal with such subjects? Do you
use an alternative contrast (i.e. effects of interest) to identify a
local maximum, are these subjects being excluded, do people usen eve
more liberal thresholds? Again, I am somewhat reluctant to exclude these
subjects, given that I believe interindividual variability to contain
lots of interesting information.
Best,
Thomas
Klaas Enno Stephan schrieb:
> Dear Thomas,
>
> This is a rather general issue that arises for any type of
> connectivity analysis, not just DCM (see Goncalves & Hall 2003,
> NeuroImage, for a discussion of this issue for SEM). The problem is
> the trade-off between obtaining time series that are, on the one hand,
> optimally characteristic for the desired effect in a specific subject
> and, on the other hand, ensuring that time series (and thus models)
> are comparable across subjects. Because the exact locations of
> activated areas (i.e. local optima in the SPMs) can vary quite
> substantially over subjects, this is not an easy task. One way of
> ensuring the comparability of extracted time series across subjects is
> to use a combination of anatomical and functional constraints. For
> example, for any area of the model, one can determine the coordinates
> of the group maximum in the random effects analysis (p<0.05 corrected)
> and use them as a starting point for searching for a nearby local
> maximum in each subject-specific data set. One can (and should)
> further require that each of these subject-specific local maxima
> should be located within the same anatomical gyrus/sulcus as the group
> maximum, but not further away than some predefined radius, e.g. twice
> the FWHM of the smoothing kernel, and to survive a threshold of p<0.05
> uncorrected. As additional anatomical criteria, one can demand that
> the activations are within a particular area (as defined by
> probabilistic atlases or functional localisers).
>
> Does this help?
>
> Best wishes,
> Klaas
>
>
>
> At 01:37 09/01/2007, you wrote:
>> Dear all,
>>
>> I am currently wondering about the logic behind the VOI extraction
>> for DCM
>> - or probably any type of connectivity analysis. The standard approach
>> seems to be that people use some straightforward contrast, i.e.
>> attention
>> vs. no attention in the classic attention to motion dataset, to identify
>> target areas in the group analyses. The same contrast is then used to
>> identify local maxima in each individual subject and use these maxima as
>> the centre coordinate for VOI extraction. However, it seems to me that
>> this approach ignores interindividual variability: what if an area A is
>> driven by some bottom-up signal from area B, but the degree of this
>> effect
>> varies across individuals? This could lead to a situation in which the
>> categorical comparison is unable to identify a local maximum for area
>> A in
>> each and every subject (well, you might guess it - I am currently
>> dealing
>> with such a dataset). Do people then use alternative contrasts to
>> search
>> for local maxima or are these subjects simply being excluded? And if the
>> latter, wouldn't we lose potentially interesting information regarding
>> interindividual variability?
>>
>> Best,
>> Thomas
>>
>> --
>> ---------------------------------------------------------
>> Dr. rer. nat. Thomas Wolbers
>> Department of Psychology
>> University of California
>> Santa Barbara, CA 93106-9660
>> U.S.A.
>> -
>> email: [log in to unmask]
>> www.psych.ucsb.edu/~wolbers
>> ----------------------------------------------------------
>
>
--
---------------------------------------------------------
Dr. rer. nat. Thomas Wolbers
Department of Psychology
University of California
Santa Barbara, CA 93106-9660
U.S.A.
-
email: [log in to unmask]
www.psych.ucsb.edu/~wolbers
----------------------------------------------------------
|