Hi,
The command line looks fine.
We did have a bug with this working for some neurologically-ordered
data. Can you run "fslorient" on the input data and tell me what the
result is?
All the best,
Mark
On 23 May 2009, at 20:28, Jeremy Elman wrote:
> Hi Mark,
>
> This is very helpful, as I had been wondering the same thing.
> However, when I ran this tool it seemed to find outlier volumes, but
> did not write an output file. Any ideas what I am doing wrong?
>
> Here is an example of my command line:
> fsl_motion_outliers /mtl/LDR/functional/LDR001-1/bold/004/f.nii 0 /
> mtl/LDR/functional/LDR001-1/bold/004/MotionOutliers.txt
>
> Thanks for your help,
> Jeremy
>
> Mark Jenkinson wrote:
>> Hi,
>>
>> There is a tool designed precisely for this.
>>
>> It is called fsl_motion_outliers and will check your motion
>> corrected data looking for points in time where there is an
>> unusual amount of residual intensity change (after motion
>> correction). Any outliers with respect to this are then
>> identified and a confound matrix created that you can
>> use in FEAT to effectively remove any changes associated
>> with these timepoints. Note that this is different from deleting
>> volumes as (i) it does not require adjusting the other model
>> EVs, and importantly, (ii) it correctly accounts for any changes
>> in signal and autocorrelation on either side of the "lost"
>> timepoint(s) as well as adjusting the degrees of freedom
>> correctly.
>>
>> To use it you just run fsl_motion_outliers on the original
>> (unfiltered and not motion corrected) data for each
>> subject/session individually. In each case it will create a
>> confound matrix which you add into the analysis for this
>> subject using the "Add additional confound EV(s)" button
>> on the "Stats" tab in FEAT. And that's it!
>>
>> Hope this sorts your problem out.
>> All the best,
>> Mark
>>
>>
>>
>>
>> On 21 May 2009, at 10:44, Klara Mareckova wrote:
>>
>>> Hello,
>>>
>>> do you happen to know if there is a relatively easy way in FSL how
>>> to
>>> idicate which slices and particular time series should be cut off
>>> from
>>> the analysis?
>>>
>>> I've analyzed the data for 50 subjects but found that they were
>>> moving
>>> a lot and therefore even if I would set quite lenient criteria and
>>> exclude everybody who moved more than 2 mm, I would end up with only
>>> 29 subjects. This is way too much and that is why I was thinking
>>> about
>>> cutting off the slices with the biggest movement (fslsplit
>>> &fslmerge).
>>> However, if I do this a problem with the time series comes out. Is
>>> there an easy way how to take care about this or do I have to go to
>>> each particular subject's design, exclude the particular time series
>>> and rerun the whole analysis?
>>>
>>> Do you also happen to have some guidelines about the exclusion
>>> criteria for motion correction? In some articles about adult
>>> participants I've seen exclusion criteria 1mm or 1degree but this
>>> seems to be too strict for my subjects.
>>>
>>> Many thanks for your help.
>>>
>>> Klara
>>>
>
|