 |
 |
Mark, et al,
The problem with going deeper down the rabbit hole is that it’s harder to make your way back out…
I tried what looked like a very nice segmentation (it performed better than any FSL FAST operations I tried) using SPM’s “new segment” and only got mediocre
results out of the BBR cost function FLIRT registration (same epi_reg line of code). I would greatly appreciate it if I could upload some data to you guys and get some more firm FSL opinions on how to get my data successfully through FSL5/FEAT6.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Eric Earl
Sent: Friday, January 17, 2014 10:36 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Mark,
Thank you for checking in. That’s very thoughtful. I am going deeper down the rabbit hole investigating white matter segmentation effects on performance (due
to the strong intensity inhomogeneity favoring intensities in the posterior of the brain). The problem still appears to stem from the point in the epi_reg script, line 365 I think, (with fieldmap on-board) where the white matter segmentation gets used, here:
$FSLDIR/bin/flirt -ref ${vrefhead} -in ${vepi} -dof 6 -cost bbr -wmseg ${vout}_fast_wmseg -init ${vout}_init.mat -omat ${vout}.mat -out ${vout}_1vol -schedule ${FSLDIR}/etc/flirtsch/bbr.sch
-echospacing ${dwell} -pedir ${pe_dir} -fieldmap ${vout}_fieldmaprads2str $wopt
I say this because I compared the
${vout}_1vol in the case of no BBR versus the case with BBR and the results were different (with
BBR the image shifted largely anterior, superior, and I think left).
I am trying a few different segmentation methods to get a better “bias field” correction and segmentation post-correction.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Mark Jenkinson
Sent: Friday, January 17, 2014 4:40 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Dear Eric,
Have you had any success?
If not, let us know and we can arrange to look at your data.
All the best,
Mark
On 13 Jan 2014, at 20:25, Eric Earl <[log in to unmask]> wrote:
Update email:
I actually got worse registration results from the GUI using the same settings and input files. Having the brain-extracted and non-brain extracted highres
image available to FEAT did not seem to help, the nonlinear registration from highres2standard was worse. What files or logs would help me to show you all these results?
Next up, I will try cleaning the data more prior to FEAT (fslroi, etc.). My masked (brain-extracted) functional, highres, and standard brains already look
good, but perhaps fslroi can remove the doubt caused by excess empty space in input volumes.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: Friday, January 10, 2014 11:56 PM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Hi,
I can't see anything obviously wrong in the images you've uploaded (especially without seeing more).
I really would strongly encourage you to try and get it working via the GUI first.
If you can't get that to work then we can see more easily what is wrong.
If the GUI run works but the other does not, then the problem to track down is related to how you are setting up the fsf file, which will be a very different problem to solve.
All the best,
Mark
On 10 Jan 2014, at 21:54, Eric Earl <[log in to unmask]> wrote:
Follow-up email…
On my Flickr page I uploaded some more FSL5 FEAT6 output images. If you have a moment, can you please look at specifically the EF_UD_shift+mag image and tell
me if that looks normal? I found it a bit unexpected considering the same image was centered in the FOV in FSL4.1 FEAT5.98.
I am still trying the different suggestions. Thanks!
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Eric Earl
Sent: Thursday, January 09, 2014 11:14 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Mark,
Thanks again for the reply. I am directly editing the FSF files and using FEAT on the command line (sort of). What I am actually doing is trying to build
a Nipype (http://nipy.sourceforge.net/nipype/) automated pipeline including FEAT in it.
I would very much like to automate the FEAT step to abstract it from our many users here and that means no GUI.
That said, I will try your suggestion in the GUI and see how it goes just to check on if that resolves the BBR error. I will also try your other suggestions
and see how they go.
P.S. Good idea with the photo sharing website, thanks! I chose Flickr and may upload images there later.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: Thursday, January 09, 2014 2:01 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Dear Eric,
Are you using the GUI or directly editing the FSF files? I strongly recommend using the GUI.
Within the GUI you provide the brain extracted image and FEAT will look for the non-brain-extracted image that shares the same name (but without a _brain in it). For example, you would enter the image myimage_brain.nii.gz and then FEAT
would look for myimage.nii.gz as the non-brain-extracted version (in the same directory). If you have not named your images in this way then you should rename them, as this is what FEAT will want. However, I suspect that this is not as much of a problem
as the other issues.
By template I did mean the atlas or standard image you are using.
If you try the suggestions that I've sent and still have trouble then let us know. We have an upload server that we will request that you send data to _if_ we think it is necessary for us to have a look. If you just want to show some
representative pictures then I'd upload some to a photo sharing website and include the link in the email.
All the best,
Mark
On 8 Jan 2014, at 19:27, Eric Earl <[log in to unmask]> wrote:
My question was actually what inputs in the FSF should receive non-brain-extracted images and what inputs should receive brain-extracted images?
Also, by “template”, do you mean the atlas I am using (standard image)?
P.S. Can anyone recommend a way or an online place for me to share my FEAT output directory? Perhaps if someone would not mind working with me over Dropbox?
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: Wednesday, January 08, 2014 12:54 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Hi,
It is worth double-checking that the segmentation is good, since BBR totally relies on this.
If that is fine then providing both brain-extracted and non-brain-extracted versions normally helps, although more for the non-linear registration to the template (assuming that you have an appropriate set of templates). As far as just
the within-subject registration is concerned, I suspect that the segmentation and the FOV (i.e. cutting off the neck) are the biggest sources of potential error.
All the best,
Mark
On 7 Jan 2014, at 22:49, Eric Earl <[log in to unmask]> wrote:
Mark, et al,
All of my images are brain extracted for the monkeys as it seems to perform better that way (usually). What inputs should I put non-brain extracted full volumes
into? In the meantime, I will look into removing neck with fslroi… Thanks some more!
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: Tuesday, January 07, 2014 12:06 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Dear Eric,
It sounds like your registration is not working well in the FSL5 (BBR) version. You should check that you are supplying the right images, both brain-extracted and non-brain-extracted versions. It can also help to remove any neck in the
images by cropping them with fslroi or robustfov. If none of that helps then let us know.
All the best,
Mark
On 5 Jan 2014, at 17:49, Eric Earl <[log in to unmask]> wrote:
Matthew,
Thank you for the helpful clarification. I did not realize that. One new question now though…
Upon further inspection, (though the registration performs better without BBR) I tried rolling back to FSL 4.1 and noticed that in the unwarping section under
the “Unwarping shift map, in voxels” the images are VERY different between FSL 5 and FSL 4.1… In FSL 4.1 the brain is centered in the field of view, but in FSL 5 it is severely out of place (very posterior in the field of view) so much so that a posterior
portion of the brain is outside the field of view. Anybody have any guesses why this might be or suggestions for things to check?
Thank you all for being so patient with my questions.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Matthew Webster
Sent: Friday, January 03, 2014 5:32 AM
To: [log in to unmask]
Subject: Re: [FSL] FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Hello Eric,
It looks like you are using B0 unwarping in your design.fsf? If that is the case then BBR registration _must_ be used.
Kind Regards
Matthew
Hi All,
I did not hear back from Stephen again and I contacted a colleague in another lab doing similar work. My colleague also said he thinks there’s a bug in FSL5/FEAT6 where BBR or 6 for the reghighres_dof always goes to BBR anyway. Can someone check into this or tell me which code/script to look at and I’ll check into this? Thank you.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Eric Earl
Sent: Friday, December 27, 2013 12:49 PM
To: [log in to unmask]
Subject: Re: [FSL] FW: FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Stephen,
Thank you so much for the fast response. So, I have tried that before and I tried it again today. When I set the reghighres_dof to 6, it switches itself back to BBR for some unknown reason (both on the command line and in the FEAT6 GUI). Here is the relevant FSF line:
# Degrees of Freedom for registration to main structural
set fmri(reghighres_dof) 6
I included the updated FSF file attached as well. Maybe see if it does the same to you?
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Stephen Smith
Sent: Friday, December 27, 2013 12:24 AM
To: [log in to unmask]
Subject: Re: [FSL] FW: FSL5 FEAT6 epi_reg problems with FLIRT BBR on monkeys
Hi - so it may be that the stuctural-based tissue segmentation that BBR needs is not working. If plain flirt is working fine, then just select that inside FEAT (eg 6DOF or 7DOF).
Cheers.
On 26 Dec 2013, at 20:42, Eric Earl <[log in to unmask]> wrote:
Dear FSL People,
I am using FEAT6 in FSL5 which is causing serious problems for my EPI (example_func) to T1 (highres) registration for rhesus macaque monkey data. I have attached my FSF file and the registration of EPI to T1 PNG image from the FEAT report. I am using nonlinear registration from highres to atlas (an appropriate monkey atlas) and a diffusion field map.
It looks to me like the problem originates in the epi_reg script, specifically on the FLIRT step that uses BBR for the cost function instead of just using mutual information or possibly correlation ratio. Testing the FLIRT command outside of the epi_reg gives a good registration with a mutual information cost function used in place of the BBR cost function.
Any help or ideas would be appreciated. Thank you all in advance.
Warm Regards,
Eric Earl, Research Assistant 2
Fair Neuroimaging Lab
Phone: 503.418.1894 Fax: 503.494.9988
Oregon Health and Science University
Department of Behavioral Neuroscience
3181 SW Sam Jackson Park Road, L470
Portland, OR 97239
<design.fsf><example_func2highres.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---------------------------------------------------------------------------