[Mrtrix-discussion] Motion correction best practise

Mon Nov 18 04:31:23 PST 2013

Hi Donald,

Thank you for all your comments.

I was not aware FSL did b-matrix update, will definitely look into this.

It's interesting that you mention we may be loosing more than we are 
gaining with motion correction, particularly with the introduction of 
artefacts and poor registration at high b-values. Is there a rule of 
thumb for when are such procedures useful/detrimental? For example, in 
fMRI movement >5mm within a single run (i.e. 5-10 minutes continuous 
scanning) is considered worrisome but salvageable and >10mm is often too 
much to be rescued by registration. Is there something like that for dMRI?

PS The Gaussian process code is now in FSL under the name EDDY ( 
http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/EDDY). I haven't personally tried 
it, but would love to hear from other users!

Kind regards,
Ivan Alvarez

PhD Candidate
Imaging and Biophysics Unit
UCL Institute of Child Health
30 Guilford Street, London, WC1N 1EH

On 18/11/13 12:15, Donald Tournier wrote:
> Hi Ivan,
>
> I'm probably not the best person to ask about this... There's a few 
> other people on this list whose opinion on the matter is much better 
> informed than mine, I cordially invite them all to chip in. ;)
>
> I've no experience with ExploreDTI, but you can ask Alex Leemans 
> directly through the ExploreDTI mailing list. That said, I'm fairly 
> confident it does the b-matrix update...
>
> As to FSL, as far as I know it also does volume-wise affine 
> registration, and it is possible to do the b-matrix update, although 
> I'm not familiar with the procedure needed to do this.
>
> I'm not familiar with any package that does slice-by-slice 
> registration, but my gut feeling on the matter is that there's 
> probably not a great deal of point to doing this as slice-wise 
> mis-registration is generally accompanied by through-plane motion, 
> which will cause signal corruption due to spin history effects. For 
> this reason, I'd consider the entire volume affected to be corrupt in 
> this case: even if there is no obvious signal drop, the chances are 
> there will be some corruption. That said, it might be worth doing if 
> your downstream processing pipeline has a way of handling outliers, etc.
>
> I'd also like to highlight that these approaches are still far from 
> perfect. I've already raised the issue on this list, but based on my 
> limited exposure to FSL's eddy_correct (which seems to be what most 
> people use), I think it often creates more problems than it solves. I 
> hasten to add that this problem may also apply to other approaches, 
> but so far I've only been exposed to data processed using 
> eddy_correct. I've come across many cases where the coregistration 
> introduces artefacts, even when the original data wasn't particularly 
> affected in the first place. These artefacts typically consist of the 
> DW images being stretched along one or more axes, probably because the 
> algorithm tries to match the parenchyma bit of the DW volumes to the 
> parenchyma+CSF parts of b=0 volume. This is particularly pronounced 
> with high b-value data, but I've also seen it happen in 
> run-of-the-mill b=1000 data too (as recently as a couple of weeks ago, 
> in fact). All this to say, if you use these tools, please don't treat 
> them as a black box, do check that they're working as expected.
>
> On the upside, Jesper Anderson recently proposed a new approach based 
> on Gaussian processes, which I think has now made it into FSL5. If any 
> other users have tried using it, feel free to comment on this...
>
> Cheers,
> Donald.
>
>
>
> On 18 November 2013 03:06, Ivan Alvarez <ivan.alvarez.11 at ucl.ac.uk 
> <mailto:ivan.alvarez.11 at ucl.ac.uk>> wrote:
>
>     Hi Donald,
>
>     I wanted to bring up motion correction again, particularly what is
>     recommended for and against in MRtrix. I am aware the issue has
>     been raised in the mailing list before, but it might be useful to
>     have an idea of what is generally a good or bad idea.
>
>     So far, I have seen people doing motion/eddy-current correction in
>     either ExploreDTI or FSL. The documentation for both is these is
>     somewhat scant and I am trying to piece together what do they
>     exactly do. This is my naive reading so far, please feel free to
>     correct me:
>
>     ExploreDTI
>     * Affine registration
>     * Whole volume at a time
>     * Updates B-matrix
>
>     FSL
>     * Affine registration
>     * Slice-by-slice
>     * Does/not/ update B-matrix
>
>     From what I understand in the discussion, the slice-by-slice
>     registration is preferable to avoid smearing artefacts across the
>     whole volume while updating the B-matrix can generally improve
>     results (http://www.ncbi.nlm.nih.gov/pubmed/19319973). Is this
>     roughly correct? If so, are there any other considerations
>     specific to MRtrix?
>
>     -- 
>     Kind regards,
>     Ivan Alvarez
>
>     PhD Candidate
>     Imaging and Biophysics Unit
>     UCL Institute of Child Health
>     30 Guilford Street, London, WC1N 1EH
>
>
>     _______________________________________________
>     Mrtrix-discussion mailing list
>     Mrtrix-discussion at www.nitrc.org
>     <mailto:Mrtrix-discussion at www.nitrc.org>
>     http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>
>
>
>
> -- 
> *Dr J-Donald Tournier (PhD)*
>
> /Senior Lecturer, //Biomedical Engineering/
> /Division of Imaging Sciences & Biomedical Engineering
> King's College London/
> /
> /
> /*A:* Department of Perinatal Imaging & Health, 1^st  Floor South 
> Wing, St Thomas' Hospital, London. SE1 7EH
> /
> /*T:* +44 (0)20 7188 7118 ext 53613/
> /*W:* 
> http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20131118/b7b64b81/attachment.html>