Hi Xuelin,
in your model you put regressors that reflect what you think will
happen in the brain as a consequence of your experiment/stimulation.
Let us say for example you have an event related paradigm where
events (very brief stimulus) occur at 20, 40 and 60 seconds after
start of experiment. You might then put in a regressor that has the
value zero for all scans/time-points, except for the time-points at
20, 40 and 60 seconds where you instead put a zero. These are often
referred to as stick-functions (for obvious reasons)
BUT, that would assume that you thought that the brain reacted
instantaneously, which we know (from experiments) that it doesn't.
Instead we know that BOLD intensity will start to rise soon after the
stimulus, reach it's highest level after ~5-6 seconds, and then start
to drop again to go negative (lower than baseline) after ~10-15
seconds after which it returns to baseline. This time course of the
BOLD response is what is known as the HRF (Hemodynamic Response
Function). Hence the stick functions above are a poor reflection of
what is going on in the brain, and therefore also a poor model. A
better model is to replace each stick with an HRF that starts at the
point of the stick.
When the stimulus function (our sticks) is no longer sticks (e.g.
slightly longer epochs) we will need to convolve the stimulus with
the HRF, but the principle remains the same.
Good luck Jesper
On 2 Apr 2007, at 02:41, Xuelin Cui wrote:
> hi Andreas:
>
> I read the slides you mentioned, the fMRI intro slides. To be
> honest to you, although it looks good, but I still dont quite get
> the experiment. Is there a paper detailedly discribe the experiment?
>
> Thanks
>
> Xuelin
>
> ****************************************
> Xuelin Cui
> Department of Electrical Engineering
> University of Hawaii-Manoa
> Honolulu HI 96822
>
> Tel: 1-808-349-0983
> Email: [log in to unmask]
> ****************************************
>
> ----- Original Message -----
> From: Andreas Bartsch <[log in to unmask]>
> Date: Saturday, March 31, 2007 10:13 pm
> Subject: [FSL] AW: [FSL] AW: [FSL] how to relate each IC to
> different tasks?
> To: [log in to unmask]
>
>> Hi Xuelin,
>>
>> why don't you start with
>> http://www.fmrib.ox.ac.uk/fslcourse/lectures/intro1/
>> mri_fmri_intro_slides.pdf , see page 49 in particular. Or join one
>> of the next FSL courses. I know - it is a long trip from Hawaii
>> but it may be worth it.
>> Lets say you have a boxcar type of experiment (see page 42 of the
>> pdf), you can model this be a series of "0" and "1" entries. You
>> can simply take this vector and correlate with the t*.txt from
>> melodic's output. Alternatively, you may want to convolve your
>> expected response vector with a hemodynamic response function -
>> e.g. a synthetic like the double-gamma HRF or even an empirical
>> one obtained specifically for your type of stimulus / brain region
>> / experiment.
>> Hope this helps-
>> Andreas
>>
>> ________________________________
>>
>> Von: FSL - FMRIB's Software Library im Auftrag von Xuelin Cui
>> Gesendet: Sa 31.03.2007 00:07
>> An: [log in to unmask]
>> Betreff: Re: [FSL] AW: [FSL] how to relate each IC to different
>> tasks?
>>
>>
>>
>> hi Andreas:
>>
>> Thanks for your reply.
>>
>> But I dont totally understand what you said. Could you please
>> explain a little bit more on that: "However, you can simply
>> correlate your design (boxcar or, for event-related studies
>> namely, your stimulus-/response-ons convolved with some type of
>> HRF)" What is a HRF? Do you have an example, or a relevant paper?
>>
>> Thanks
>>
>> Xuelin
>>
>> ****************************************
>> Xuelin Cui
>> Department of Electrical Engineering
>> University of Hawaii-Manoa
>> Honolulu HI 96822
>>
>> Tel: 1-808-349-0983
>> Email: [log in to unmask]
>> ****************************************
>>
>> ----- Original Message -----
>> From: Andreas Bartsch <[log in to unmask]>
>> Date: Tuesday, March 20, 2007 3:26 pm
>> Subject: [FSL] AW: [FSL] how to relate each IC to different tasks?
>> To: [log in to unmask]
>>
>>> Hi,
>>> Strictly speaking: you can't (- at least not by just running a
>>> single experiment and analysing it;)
>>> However, you can simply correlate your design (boxcar or, for
>>> event-related studies namely, your stimulus-/response-ons
>>> convolved with some type of HRF) with the time-course (in SD
>>> units) of the respective tXX.txt file. Or even easier, if you have
>>> no spatial hypothesis whatsoever, select those Ics whose FFT are
>>> peaking at the appropriate frequency and then look at the time-
>>> courses, if they are synchronous to your paradigm.
>>> Hope that helps-
>>> Andreas
>>>
>>> -----Ursprüngliche Nachricht-----
>>> Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im
>>> Auftrag von Xuelin Cui
>>> Gesendet: Dienstag, 20. März 2007 00:55
>>> An: [log in to unmask]
>>> Betreff: [FSL] how to relate each IC to different tasks?
>>>
>>> hi folks:
>>>
>>> I here have a question. I am doing the ICA analysis on fMRI data.
>>> But I am confused on how to relate rach IC extracted from the fMRI
>>> data to different tasks. Put in this way: if I see an IC, how can
>>> I tell the IC is caused by what kind of reason?
>>>
>>> Anyone could give any suggestions? I really appreciate your
>> thoughts.>
>>> Thanks
>>>
>>> Xuelin
>>>
>>> ****************************************
>>> Xuelin Cui
>>> Department of Electrical Engineering
>>> University of Hawaii-Manoa
>>> Honolulu HI 96822
>>>
>>> Tel: 1-808-349-0983
>>> Email: [log in to unmask]
>>> ****************************************
>>>
>>
|