Hi Steve (and Joe),
Thanks for looking into this and helping out!
I've never played around with SUSAN, so I'm not exactly sure how it works. I
will look into this and see what happens. If it achieves the same quality of
masks as FSL 3 (i.e., no holes in the brain), then I suspect it should work
nicely. I think avoiding the holes in the center of the brain and getting a
good mask at the third level are the top priorities (the third level mask was
where this "problem" was clearly the most noticeable). For now, the workaround
I posted earlier has worked very well for me and others at my institution, but
it does have the same drawback as the earlier version of FSL 3 (i.e.,
voxels on
the edge of the brain will get blurred into non-brain voxels). It's only a
couple of extra lines of code, too.
I still wonder about how much registration is affected by making the mask a
little too big (Joe also mentioned this). I also haven't specifically tested
this yet, but it seem like the "halo-like rim" could get treated as the
edge of
the brain and throw off the scaling when putting the functional data and cope
images in standard space. Sorry if this is not the case (I know you said it
shouldn't be in one of your earlier posts). I guess it's not completely clear
why the scaling wouldn't get thrown off if the edges of the cope images are
outside of the actual brain and consequently get treated as the edge of the of
the brain when registering to the standard template with 12 DOF.
Thanks again for all the help and clarifications! I really appreciate it!
Cheers,
David
Quoting Steve Smith <[log in to unmask]>:
> Hi Joe, David, et al.,
>
> Joe's kindly sent us some data to play with and we've had a detailed
> look. The difference between FSL 3 & 4 in the FEAT brain masking is
> not primarily relating to changes in the brain-extraction or even the
> thresholding details. The difference is that in FSL4 the spatial
> smoothing (which is applied after the brain extraction and the
> intensity thresholding) is only applied _within_ the current brain
> mask and does not blur out the effective brain mask in the same way
> that it used to. The effect is that the new brain mask definition is
> more "accurate" than it was before, and is in general about 1 voxel
> eroded compared with FSL3 brain masks (where the spatial smoothing
> effectively dilated the brain mask).
>
> So - what do people want? More "accurate" brain masking or more
> liberal? We could just dilate the brain mask further to start with,
> though that would then lose the advantage of only applying the
> spatial smoothing within the brain which we currently have. A
> solution to this would be to use SUSAN smoothing instead of Gaussian
> to get the best of both worlds - if we set the SUSAN intensity
> threshold to say range/10 then you'd get effectively Gaussian
> smoothing within the brain, no blurring of the brain edge, and still
> could include a rim of non-brain voxels around the edge of the
> brain. What do you think?
>
> Cheers.
>
>
>
>
> On 12 Mar 2008, at 15:25, Joseph T. Devlin wrote:
>
>> Hi Steve,
>>
>> I took the opportunity to go back through some data I collected at
>> FMRIB and re-analysed using the new (FSL 4) brain masking and found
>> the same problems I'm seeing in my current data. That is, the
>> analyses using FSL 3.3 produced nice results on my FMRIB data but
>> looked much poorer when analysed in 4.0 (just the masking -- the
>> rest was fine). This seems consistent with David's experience, if I
>> read his earlier posting correctly.
>>
>> For what it's worth, I got one other email off list noting they
>> preferred the old method more too.
>>
>> So I'd also be curious to know whether anyone else has noticed this
>> behaviour with their masking?
>>
>> Many thanks
>> Joe
>>
>
>
> ---------------------------------------------------------------------------
> 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
> ---------------------------------------------------------------------------
>
--------------------------------------------------
David V. Smith
Graduate Student, Huettel Lab
Center for Cognitive Neuroscience
Duke University
Durham, NC 27708
www.mind.duke.edu
--------------------------------------------------
|