Hi Katie,
There aren't generally hard rules on these things. I would guess that
if you have just a few voxels giving the warning (as you have here)
then you probably shouldn't worry. When you look at the cope and
varcope timeseries at these voxels, if it's not obvious where the
outliers are, and when you look at the relevant subjects' images
around this voxel, nothing appears badly corrupted (etc) then you're
probably ok.
The warning happens when FLAME stage 2 is sufficiently different from
stage 1, meaning that the more accurate stage 2 MCMC is different
from the slightly less accurate FLAME stage 1 - i.e. that the
approximations behind the stage 1 estimation are less valid than
usual. However, for just a few voxels, I think you probably shouldn't
worry.
Oh for the good-old-days when everyone used to use just OLS
estimation and such issues were hidden from view :)
Cheers, Steve.
On 28 Oct 2006, at 17:41, Katie Karlsgodt wrote:
> Hi,
> Ok, thanks. I think the problem is that I'm not totally clear on
> what the
> practical implications of this are. I got the warning at 5 voxels
> for the
> cope I ran... at each voxel, a different subject (or 2) seems to be
> the
> outlier, some of the voxels are in areas I'm interested in. Is this
> something that I would throw away subjects based on? or, would that
> only
> happen if there were a certain number of voxels at which the warning
> appears? If I flip through the var_filtered_func_data and see
> people who
> look different to my eye, is there some way of determining if they are
> enough of an 'outlier' to eliminate?
> And, if it is not something to eliminate people on, does it mean I
> should be wary of interpreting the results around the locations of
> those
> voxels? Does the warning indicate that flame was unable to do stats
> properly, or just that I should be aware that there are a few
> places with a
> lot of variance? Sorry for all the questions, I'm just not
> completely sure
> how to proceed from here.
> thanks,
> Katie
> ___________________________________
> Katie Karlsgodt
> Dept of Psychology/Cognitive Neuroscience
> University of California, Los Angeles
>
> [log in to unmask]
> phone: (310) 794-9673
> fax: (310) 794-9740
>
>
>
>> From: Steve Smith <[log in to unmask]>
>> Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
>> Date: Sat, 28 Oct 2006 08:02:43 +0100
>> To: <[log in to unmask]>
>> Subject: Re: [FSL] Flame stage 1+2 error
>>
>> Hi - the log output will also tell you _where_ those voxels are - so
>> you can open a new FSLView, load in the filtered_func_data and
>> var_filtered_func_data, and go to this voxel, and turn on the
>> timeseries display for both 4D datasets - then this may show you
>> where and who are the problems.
>>
>> Hope this helps?
>>
>> Cheers, Steve.
>>
>>
>> On 27 Oct 2006, at 20:05, Katie Karlsgodt wrote:
>>
>>> Hello,
>>> I'm running a group of 31 subjects (2 groups of 14 and 17) in
>>> Flame and
>>> am getting the :"WARNING: FLAME stage 2 has given and abnormally
>>> large
>>> difference to stage 1" error at 5 different places in the analysis.
>>> I see
>>> from the responses in the archives that this may be due to
>>> variance/outliers:
>>>
>>> "this normally means that you have strong outliers in the data,
>>> particularly as your number of inputs is not small. You should look
>>> at the 4D cope and varcope files input to the higher-level analysis
>>> and flick through the timepoints to see if you have any strong
>>> outliers - that may resolve this."
>>>
>>> I've flipped through, and I do seem to have some people who look
>>> different,
>>> I'm just not sure how to actually define an outlier in this case?
>>> There was
>>> one subject in particular who was really much different, so I
>>> removed him
>>> and reran it, but still got the error. Do you have a recommended
>>> way for
>>> checking these images? I'm just not sure what I'm looking for.
>>>
>>> thanks,
>>> Katie
>>> ___________________________________
>>> Katie Karlsgodt
>>> Dept of Psychology/Cognitive Neuroscience
>>> University of California, Los Angeles
>>>
>>> [log in to unmask]
>>> phone: (310) 794-9673
>>> fax: (310) 794-9740
>>
>>
>> ---------------------------------------------------------------------
>> ---
>> ---
>> 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
>> ---------------------------------------------------------------------
>> ---
>> ---
>>
>>
>>
------------------------------------------------------------------------
---
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
------------------------------------------------------------------------
---
|