Ah - the normalization argument makes a lot of sense!
We will try it again with that setting turned off - I wish we always
had this problem of too much activation...
Thanks very much for this insightful reply!
Rick
On Jul 22, 2004, at 6:35 PM, Stephen Smith wrote:
> Hi Rick - thanks for sending the data.
>
> The group-level analysis looks fine (though it would normally be hard
> to
> find mixed-effects activation with only 4 lower-level analyses - but
> your
> activation is huge). I checked with OLS as well as FLAME and you still
> get
> the negative background values.
>
> The first-level analyses also look fine, in the sense that it's not
> artefacts in the data causing the problem (note that MELODIC PICA
> analysis
> did show a few spikes still left in btw, but I don't think that's the
> problem).
>
> I'm pretty sure I know what the problem is - though without the
> original
> data I can't check to be certain. You have used intensity
> normalisation,
> which is normally turned off by default, for various reasons including
> this one: your data is so lovely, and the activation so huge, that
> intensity normalisation is interacting with the strong activation to
> produce "negative activation" in the background (nonactivated) voxels,
> for
> obvious reasons. I'm pretty sure this is what's going on - you can
> easily
> re-run without intnorm and see if the problem improves.
>
> A much less likely possibility is that structured noise (eg resting
> state
> networks) is causing the problem, but I'd try the above first.
>
> Let us know how you get on! Cheers, Steve.
>
>
> On Tue, 20 Jul 2004, Rick Hoge wrote:
>
>> We've been analyzing some block design data using FEAT, and have
>> encounted a strange negative
>> bias in the Z and T images generated for our groups by FLAME. We do
>> see a set of positive foci in
>> plausible locations, but these appear against a background of a
>> consistently negative Z score of
>> around -5. This would suggest there's either something wrong with
>> our data (but these look ok
>> from what we can tell) or, more likely, that we are setting up the
>> analysis incorrectly.
>>
>> I am wondering if this symptom calls to mind any kind of obvious
>> misconfiguration, or if anyone
>> has advice on how to debug this sort of thing. We did look at the
>> cope images, and these also
>> appear to have the negative bias in much of the brain.
>>
>> The design contains three stimulus types (EV's) in non-overlapping
>> blocks, and the problem is
>> appearing in the general contrast testing for presence of response to
>> all three stim types. The F
>> images show a global background of abnormally high value, and you can
>> see the zero crossing
>> separating most of the brain (negative) from the isolated positive
>> foci.
>>
>> Any advice on how to resolve this would be much appreciated - I can
>> provide all or part of the .fsf
>> file or output images if needed.
>>
>> Rick
>>
>
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
|