Hi Steve,

Thanks for the reply -- sorry I was bit unclear and short on details. One of the ICA maps onto a region that was associated with both task conditions. The key distinction is the addition of this one other region that is "negative" in the group ICA -- my GLM analysis in FEAT did not pick that up. 

It seems reasonable that both regions (positive and negative) could simultaneously contribute to the observed association with the task conditions. However, I'm not 100% sure if I have evidence for that claim based on the observation that the dr_stage1 time courses for this component are associated with both conditions. 

Could I split this map into positive and negative maps, re-estimate the dual regression to obtain new dr_stage1 time courses, and then regress the new dr_stage1 time courses onto my task again? Would that allow me to quantify how the positive and negative parts of the component contribute to my task conditions? (I've seen what appears to be a similar approach at the end of the Results section of a JNeurosci paper from Leech and colleagues: http://www.jneurosci.org/content/31/9/3217.long )

Thanks!
David



On Jan 30, 2015, at 5:52 AM, Stephen Smith <[log in to unmask]> wrote:

Hi David

If I understand you correctly, it seems that the group-ICA spatial-maps might not necessarily map well onto distinct areas of interest wrt the task activation?  I'm not quite sure I follow the details.....:


On 30 Jan 2015, at 05:23, David V. Smith <[log in to unmask]> wrote:
Hello,
I'm having trouble interpreting how my independent component maps relate to my task.

I used the dual_regression script to obtain subject-specific time courses for all the components in my group ICA. The component/network I care about in the group ICA contains voxels that are significantly negative and voxels that are significantly positive (in roughly equal proportions). When I regress the dr_stage1 time courses onto my task design matrices (which have two main conditions), the positive/negative component of interest is significantly associated with both conditions, but the association is much stronger for one condition compared to the other condition.

...right - don't forget that when you have multiple conditions, there's no guarantee (or necessarily even expecation) that ICA will separate out components according to those different conditions/contrasts - as they are likely to be spatially quite overlapping if they are similar tasks.

You might want to start by looking at how the ICA maps relate to maps from FEAT GLM analysis?

Cheers.



It seems like the difference I'm observing could arise due to how the conditions modulate a) the positive portion of the component, b) the negative portion of the component, or c) both the positive and negative portion of the component. Is that correct? If so, what sort of tests could I run to distinguish between these alternatives?

Thanks!
David


---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------

Stop the cultural destruction of Tibet