Hi again Mark and fsl list,
I am writing on a topic closely related to a question I once asked (quoted
below).
I understand that, in fsl, whenever one models two EVs (EV1 & EV2), a
contrast [1 0] for EV1 reflects the variance associated with EV1 over and
above the variance accounted for by EV2. In other words, I understand that
is mathematically equivalent to a within-subjects residualization of the
effects of EV2 from EV1.
My question, on the other hand, regards a between-subjects residualization
procedure. Imagine that, in an investigation of individual differences,
participants that have the largest changes in a number of brain areas during
a period model by EV1 may also have the largest changes in those areas
during the period modeled by EV2. I'd like to run a voxelwise regression of
a behavioral variable on the EV1 statistical maps, while controlling for
(residualizing) any individual differences of brain activation on EV2 maps.
In one sentence, the question is: is a *within-subjects* residualization
procedure (routinely part of fsl) sufficient to control for individual
differences in EV2 statistical maps when running a voxelwise regression
*across subjects* on EV1 statistical maps?
Thanks much for any clarification or recommendations,
Regina
On Wed, 29 Apr 2009 09:55:46 +0100, Mark Woolrich <[log in to unmask]>
wrote:
>Hi Regina,
>
>It sounds like you want to know how, within a voxelwise GLM containing
>2 EVs, you can find only the explanatory power associated with (i.e.
>variance explained by) EV1 above and beyond EV2 and vice versa. The
>answer is that when you do [1 0] or [0 1] contrasts the GLM will
>automatically take care of this for you. For example, if the EVs are
>partially correlated then the GLM fitting of the parameter estimate
>for EV1is only driven by the component of EV1 that is orthogonal with
>(uncorrelated to) EV2 - and the resulting statistics will
>automatically reflect this accordingly.
>
>Cheers, Mark.
>
>----
>Dr Mark Woolrich
>EPSRC Advanced Research Fellow University Research Lecturer
>
>Oxford University Centre for Functional MRI of the Brain (FMRIB),
>John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
>
>Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich
>
>
>
>
>On 29 Apr 2009, at 02:45, Regina Lapate wrote:
>
>> Dear all,
>>
>> I have a question that I can imagine two potential ways of answering
>> it, and
>> while I am comfortable with how to implement it the first way, I
>> would like
>> to know if there is a practical way to implement it the second way
>> (in fsl).
>> The question is:
>>
>> - If one wants to residualize brain activation associated with
>> condition A,
>> from brain activation associated in condition B...
>>
>> 1) In a ROI approach, I assume one can extract parameter estimates
>> associated with BOLD changes due to condition A (cope1), similarly for
>> condition B(cope2), and regress cope1 on cope2 using a statistical
>> program
>> while saving the residuals of this regression;
>>
>> 2) But what about in a voxel-wise GLM approach in fsl, where both
>> conditions
>> A and B are EVs in the model? Is there a practical way where fsl can
>> save/output the residuals that remain after accounting for the
>> variance
>> associated with condition A, so that I could look at the variance B
>> accounts for once variance due to A has been removed?
>>
>> Thanks much for any suggestions!
>>
>> Regina
>>
|