Hi Brad and all,
I have a suggestion to make. If you have a look at a paper by
Friston et al entitled "Movement-related Effects in fMRI Time Series"
that may help you out. The paper describes a method of including 24
regressors that account for residual motion that built-in motion
correction paradigms, like McFLIRT, may not be able to remove. I
don't know how it will handle jerky movements, but it may help.
Brad, I'm sure you would be interested to know that I use this
method with the fMRI measurements that I have.
Todd Penney
Quoting Bradley Goodyear <[log in to unmask]>:
> Hi Pete.
>
> One thing to keep in mind is that if your experiment is related to the
> resting-state of motor-related areas, then the correction of a
> significant head motion in this way may not solve all of your problems.
>
> A head motion like that will significantly modulate the hemodynamic
> response in motor areas that will extend in time beyond the time of
> motion, and may be evident in the time course.
> Hence, you may not have a true resting state anymore. At least this has
> been our experience. We have tried to use an additional regressor to
> model the hemodynamic response to the head motion, but have had little
> success. We found that it's best to collect several runs and discard
> any with motion artifacts.
>
> -Brad
>
> On Jul 5, 2007, at 10:14 AM, Doug Greve wrote:
>
>> Pete,
>>
>> The right way to remove a time point is to add a regressor (custom
>> 1-column) with 1 at that time point and 0 everywhere else. Do NOT
>> convolve with a shape or add a deriviative. DO apply temporal filtering
>> if you are using it on your data. If you have a 2nd time point to
>> remove, then add another regressor (don't put a 2nd 1 in your 1st
>> regregressor).
>>
>> doug
>>
>>
>>
>>
>> On Thu, 5 Jul 2007, Peter Fried wrote:
>>
>>> Hi,
>>>
>>> I have a question about correcting for motion in resting-state fmri data.
>>>
>>> We collect 180 volumes (TR=2s) of fmri data during a resting-state. The
>>> data is first processed using the pre-stats tab under FEAT. The parameters
>>> we use are:
>>>
>>> Motion Correction (mcflirt)
>>> B0 unwarping
>>> Slice-timing correction
>>> Bet brain extraction
>>> Spatial smoothing FWHM = 6 mm
>>> Intensity Normalisation
>>> Highpass temporal filtering
>>>
>>> plus, Melodic
>>>
>>> Here's the problem: MCFLIRT seems to do a good job when there is gradual
>>> movement over the course of the scan. However, if there is a sudden
>>> movement (say, the subject coughed or jerked their arm), MCLFIRT does not
>>> seem to be able to correct it. To make matters worse, there are often
>>> motion spikes that show up in the timecourse of non-motion components.
>>>
>>> We've tried filtering out some of the worst melodic components, but that
>>> doesn't seem to effect components that have been (for lack of a better
>>> term) corrupted by motion. The best results we've had have been from
>>> splitting the 4D volume, removing the volumes with the most motion and
>>> merging the remaining volumes back together. Even though we don't have a
>>> block design, I'm still wary of doing this, especially if there is a
>>> better way.
>>>
>>> Here's my question: Is there any way to do a more robust elimination of
>>> motion, other than the parameters I've mentioned above? Also, should
>>> deleting volumes be avoided at all costs or is it okay to remove a few
>>> with bad motion?
>>>
>>> Thanks.
>>> -Pete
>>>
>>>
>>>
>>
>> --
>> Douglas N. Greve, Ph.D.
>> MGH-NMR Center
>> [log in to unmask]
>> Phone Number: 617-724-2358 Fax: 617-726-7422
>>
>> In order to help us help you, please follow the steps in:
>> surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
>>
>>
|