Thanks for the help, Steve. I've uploaded the data and the ref # is 530572.
Since prestats was run in MELODIC, I'm including one of those directories. I'm
also including a mask from the second level to give an indication of how the
problem gets worse as you get further down the analysis pipeline.
My guess is that the edges of the mask (which extend past the actual brain due
to the low setting on the brain/background threshold) eventually get
treated as
the edge of the brain because of normalization (but seemingly only at
the third
level). They generally look OK in the second level in that they extend
past the
brain, but they're still very jagged and rough around the edges and look
dramtically worse at the third level. However, I have no idea why the lowest
setting on the %brain/background setting would still erode voxels in
the middle
of the brain, which is clear in some of the images I've sent.
Thanks again,
-David
Quoting Steve Smith <[log in to unmask]>:
> Hi David,
>
> It does seem like the mask-generation is being over-conservative
> (i.e. under-inclusive) and this is certainly not what we generally
> see (as you rightly say the default behaviour aims to be
> over-inclusive). I think the thing to do is to work out why the
> masks are including this many holes and erosions at the brain edge.
> I would start by identifying a single first-level session that
> contains what you consider to be suboptimal masking, and work out
> what stage of the preprocessing is causing this, and why the default
> %thresh value is not giving you what you want. If you'd like to
> upload a complete first- level FEAT directory, plus the original
> input 4D data, we can easily check into this.
>
> Cheers, Steve.
>
>
>
>
> On 27 Nov 2007, at 19:09, David V. Smith wrote:
>
>> Hi Steve,
>>
>> The smoothing argument makes sense and this would seemingly be an
>> improvement
>> over the older version. However, the extreme edges of the mask are
>> supposed to
>> go beyond actual brain voxels (i.e., the mask is supposed to be
>> overinclusive),
>> so blurring these non brain voxels with the background shouldn't
>> really matter
>> unless the smoothing kernel is huge (I think).
>>
>> In any case, the end result is a mask that tends to exclude voxels
>> that I would
>> like to keep. For example, please see the attached image of a 3rd
>> level mask
>> that was derived from first level feats that all had a reduced
>> brain/ non-brain
>> threshold of 1%. This is certainly an improvement over using the default
>> setting (which I posted about back in October), but it isn't perfect,
>> especially in comparison to the older version of FSL. OFC, parietal
>> cortex and
>> other areas have holes in them.
>>
>> As a fix, could I simply run all of the preprocessing steps from the
>> command
>> line and do the smoothing before BET? I suspect all of the typical options
>> within the pre-stats tab have easily scriptable programs (the only
>> one I'm not
>> sure about is the highpass filter).
>>
>> Thanks,
>> David
>>
>> --------------------------------------------------
>> David V. Smith
>> Graduate Student, Huettel Lab
>> Center for Cognitive Neuroscience
>> Duke University
>> Durham, NC 27708
>> www.mind.duke.edu
>> --------------------------------------------------
>>
>>
>> Quoting Steve Smith <[log in to unmask]>:
>>
>>> Hi - yes, this is expected I think, and in general is due to an
>>> "improvement": in the old version, the spatial smoothing was
>>> just applied homogeneously across the whole image, including the
>>> effect of blurring the edges of the brain into the background. In
>>> FSL4.0 this is improved - the smoothing kernel is limited to lying
>>> only within the brain mask area, so that non-brain voxels are not
>>> included into the smoothing of near-brain-edge voxels. I think
>>> that this in general a better thing to do - one can't argue that
>>> the non-brain voxels have any useful information to add to the
>>> near-brain-edge voxels - and explains what you're seeing here. If
>>> you really think that this is ending up excluding voxels that you
>>> want kept in (which I suspect it's not), then yes you can just
>>> reduce the brain/non-brain threshold value if you prefer that.
>>>
>>> Hope this makes sense? Cheers, Steve.
>>>
>>> p.s., Note that none of the above affects head motion correction
>>> or registration, both of which effectively happen before spatial
>>> smoothing.
>>>
>>>
>>>
>>>
>>> On 27 Nov 2007, at 05:17, David V. Smith wrote:
>>>
>>>>
>>>> Hi,
>>>>
>>>> I've had trouble with the quality of the masks that I'm getting
>>>> from the new
>>>> version of FSL; however, the older versions of FSL seem to make
>>>> masks that are
>>>> perfectly fine. See attached image.
>>>>
>>>> What's changed with the new version of FSL, and how can I make it
>>>> produce masks
>>>> like the older version? These differences are not subtle and they
>>>> led to the
>>>> significant problems I posted about earlier
>>>> (http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0710&L=FSL&P=R68794&D=0&H=0&I=-3&O=T&T=0
>>>> ).
>>>> I've tested this on data from different scanners and I don't
>>>> think the problem
>>>> is because of data quality or "weird intensities" as Steve
>>>> suggested earlier.
>>>>
>>>> Reducing the %Brain/Background setting to 1 has helped a little,
>>>> but I was
>>>> hoping you guys could tell me if anything is wrong with the new
>>>> version and if
>>>> something is wrong, how can it be fixed.
>>>>
>>>> Thanks,
>>>> David
>>>> <fsl_mask_comparison.png>
>>>
>>>
>>> ---------------------------------------------------------------------------
>>> 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
>>> ---------------------------------------------------------------------------
>>>
>>
>>
>> <3rdlvl_mask_1percent.png>
>
>
> ---------------------------------------------------------------------------
> 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
> ---------------------------------------------------------------------------
>
|
