Hi,

Steve was right about the padding issue here. If you add “--paddingsize=5” to the applywarp commands, the issue goes away. 

applywarp -i reg/unwarp/FM_UD_fmap_mag_brain_mask -r example_func --paddingsize=5 --rel --premat=reg/unwarp/FM_UD_fmap_mag_brain2str.mat --postmat=reg/highres2example_func.mat -o reg/unwarp/EF_UD_fmap_mag_brain_mask
for v in `seq 0 255`; do
v=`zeropad ${v} 4`
applywarp -i grot${v} -o grot${v} --premat=mc/prefiltered_func_data_mcf.mat/MAT_${v} -w reg/example_func2highres_warp.nii.gz -r example_func --paddingsize=5 --rel --postmat=reg/highres2example_func.mat
done

Cheers,
David


On Nov 5, 2014, at 7:12 PM, David V. Smith <[log in to unmask]> wrote:

Hi Steve,

Thanks for the feedback. I went back and looked at an old analysis from a few years ago (with FSL 4.1.8), and it didn’t have the same sort of noise component following B0 unwarping (even with partial brain coverage). I know a lot of things have changed with B0 unwarping since 4.1.8, but maybe the padding issue you mentioned is one of them. It seems a couple of other people have noticed this issue since the release of FSL 5:

JISCMail - FSL Archive - Melodic denoising and b0 unwarping problem (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;f16fd92d.1306)

JISCMail - FSL Archive - MELODIC denoising with b0 unwarping (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;630103a6.1301)

I’m still wondering if this issue is strictly cosmetic. These components do load onto other areas — albeit to a much lesser degree than the edge slices. In addition, these components dominate the ICA, accounting for about 50% of the total explained variance; this results in many components that are split into subcomponents when B0 unwarping isn’t applied (and it seems like this feature would complicate denoising in FIX).

Thanks again!

David






On Nov 5, 2014, at 11:13 AM, Stephen Smith <[log in to unmask]> wrote:

Hi David

Yes - I think this is just because the full brain was not quite acquired (ie bottom of cerebellum cut off) - and this effect is where the bottom slices are varying in exactly what is happening wrt the motion correction (etc).   The fact that some ICA components are finding these effects should not affect other ICA component estimation within the main body of the brain, which should hopefully be independent of that (as it were ;-)

Having said this, I seem to remember that we used to use extra slice "padding" when interpolating and carrying out mcflirt motion correction, to reduce such "cosmetic" end-slice effects.....  it's possible that the latest pipeline that combines motion correction with other things might not be doing such padding - Matthew, could you check on that?

Thanks, Steve.







On 5 Nov 2014, at 15:39, David V. Smith <[log in to unmask]> wrote:

Thanks Steve and Matt. 

I’m not concatenating across runs (these are just single-session ICAs). 

Here’s a link to some example output (with large image files removed; the screen shot Steve requested is also in the archive):

I think Steve is right. The artifact is most prevalent near the edges (particularly the bottom two slices). But there are suprathreshold voxels in other parts of the brain, some of which appear near the striatum/ventricles (e.g., components #3 and #10). 

Maybe I’m messing something up with the fieldmap? Either way, I’m concerned about this artifact affecting the ICA and contributing noise to other parts of the brain. 

I’m also wondering if my preprocessing steps are optimal for an interleaved sequence: motion correction / unwarping followed by slice-timing correction. 

Thanks again for your help!

Cheers,
David



On Nov 5, 2014, at 12:30 AM, Matt Glasser <[log in to unmask]> wrote:

Please also do what Steve suggests about showing example components.  I would not expect field map correction to cause artifact around internal tissue boundaries away from the edges.  Also, the field map being applied should be an entirely static effect (i.e. whether or not you apply a field map the motion related artifacts should not change).  

Are you concatenating across runs or anything like that?

Peace,

Matt.

From: "David V. Smith" <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Tuesday, November 4, 2014 at 10:47 PM
To: <[log in to unmask]>
Subject: Re: [FSL] noise components following B0 unwarping

Sure. Our coverage is OK, but there’s definitely not much room to spare (37 slices; slice thickness = 3.3mm). Note the “artifact” also appears around tissue boundaries (e.g., ventricles and caudate); i.e., it’s not strictly limited to the top/bottom slices, although it is most notable there. 

Thanks,
David




<Screen Shot 2014-11-04 at 11.39.17 PM.png>
On Nov 4, 2014, at 10:37 PM, Matt Glasser <[log in to unmask]> wrote:

Can you post some images showing what kind of coverage you have?  I’ve certainly never seen anything like this, but the data I’ve worked with cover the whole brain with room to spare and the only place where anything like this might occur would be in the lower brain stem.  

Peace,

Matt.

From: "David V. Smith" <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Tuesday, November 4, 2014 at 8:54 PM
To: <[log in to unmask]>
Subject: Re: [FSL] noise components following B0 unwarping

Same issue without slice-timing correction. And it comes up (to approximately the same degree) in subjects who move a lot and in subjects who move a little. 

It also happens with a different sequence using a “regular up” slice acquisition instead of interleaved. 

The key variable seems to be the field map correction, which does a nice job of correcting spatial distortions but at the cost of introducing these noise components. 

Thanks,
David


On Nov 4, 2014, at 8:56 PM, Matt Glasser <[log in to unmask]> wrote:

What happens if you don’t do slice timing correction?  Is your subject moving a lot when this happens?

Peace,

Matt.

From: "David V. Smith" <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Tuesday, November 4, 2014 at 1:52 PM
To: <[log in to unmask]>
Subject: Re: [FSL] noise components following B0 unwarping

Hi Matt,

I’m using the default order implemented in FEAT. 

So, motion correction matrices are estimated first:
/usr/share/fsl/5.0/bin/mcflirt -in prefiltered_func_data -out prefiltered_func_data_mcf -mats -plots -reffile reg/unwarp/EF_D_example_func -rmsrel -rmsabs

Those transforms are then applied during unwarping:
/usr/share/fsl/5.0/bin/applywarp -i grot0000 -o grot0000 --premat=mc/prefiltered_func_data_mcf.mat/MAT_0000 -w reg/example_func2highres_warp.nii.gz -r example_func --rel --postmat=reg/highres2example_func.mat

And slice-timing correction is performed after unwarping:
/usr/share/fsl/5.0/bin/slicetimer -i prefiltered_func_data_unwarp --out=prefiltered_func_data_st -r 2.000000 --odd

Is this default order optimal, especially given an interleaved slice acquisition? The MRC imaging wiki suggests that, for an interleaved slice acquisition, slice-timing correction should probably occur before motion correction. 

Thanks for any guidance!

Cheers,
David



On Nov 4, 2014, at 11:20 AM, Matt Glasser <[log in to unmask]> wrote:

Are you doing motion correction before B0 unwarping?

Peace,

Matt.

From: "David V. Smith" <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Tuesday, November 4, 2014 at 9:58 AM
To: <[log in to unmask]>
Subject: [FSL] noise components following B0 unwarping

Hi,

I am planning to use FIX to denoise my data. I’ve noticed that I get very noisy components when I include B0 unwarping in my preprocessing. These components have a distinct time course (see attached), and they seem to load mostly on the top/bottom two slices and around tissue boundaries). These noise components are explaining up to 50% of the total variance in my data (according to the MELODIC output).

Have other people seen this type of artifact following B0 unwarping? Are these components reflecting an interaction between head motion and the distortions? 

Thanks!
David

<t1.png>
<t1.png>

----------------------------------------------
David V. Smith, Ph.D.
Postdoctoral Fellow, Delgado Lab
Department of Psychology
Rutgers University
Newark, NJ 07102
----------------------------------------------



<Screen Shot 2014-11-04 at 11.39.17 PM.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
---------------------------------------------------------------------------

Stop the cultural destruction of Tibet