[Mrtrix-discussion] Crossing-fibers gray matter CSD

[Donald Tournier]

Hi Wim,

I don't see anything wrong with your preprocessing, everything seems
OK. The only thing I might recommend is to drop the lmax to 6 for the
estimate_response step - at b~1200, there really won't be much in the
l=8 term. But then, the response looks OK, the encoding is fine, so I
don't think it would make any difference to your results.

The only thing I can recommend is to have a look at the (very sparse!)
literature on cortical tracking with multi-fibre algorithms. The only
references I can think of are a few ISMRM 2005 abstracts by the MGH
group, in particular pages 393, 583 & 584. There are a few subtle
remarks about their observation of 'horizontal' fibres within the
cortex, but nothing particularly overt... You might also find
something of interest in their NeuroImage 41: 1267-1277 (2008) paper.
There might be other published work on the topic, but I haven't come
across them yet.

In short, I can't see anything wrong with the analysis as such, but
your results are so strange-looking that I would be very reluctant to
trust them without some robust corroborating evidence. The effect (in
terms of the consistent 'horizontal' fibre orientations within the
cortex) could very well be real, although the long range connections
that you show are probably an artefact of the tracking (the algorithm
simply doesn't know when to stop).

You might also think about trying it with a higher b-value. Around
b~3000 should give you a much crisper response function, and hence
more reliable fibre orientations.

Not sure all this helps you much, but I'm afraid that's as much as I can do...
Cheers,

Donald.


On Mon, Apr 13, 2009 at 10:37 PM, Wim Otte <wim at invivonmr.uu.nl> wrote:
> Hi Donald,
>
> Indeed strange tracking results... I attached several files:
> - screenshot: the first and second principal eigenvectors, constructed
> with FSL, overlaid on the FA-map ('ground truth'...).
> - screenshot: response function estimated in white matter voxels
> - encoding.txt: (constructed from bvals and bvecs using the script
> combine_bvecs_bvals.pl)
> - csd_steps.sh: script used to construct fibers from the gray-matter seed.
>
> Great that you are willing to check these files!
>
> Wim
>
>
>
> On Mon, Apr 13, 2009 at 4:29 AM, Donald Tournier
> <d.tournier at brain.org.au> wrote:
>> HI Wim,
>>
>> These are some of the strangest tracking results I have seen so far!
>> On the other hand, it looks like you are tracking through cortex, so
>> maybe that could explain a few things. The main orientations look OK,
>> even in the cortex (I assume this is a very young animal). I'm
>> surprised that the non-radial directions in the cortex are so
>> consistent - they all seem to be pointing at 45°. Not too surprised
>> that the tracking down to the white matter stops so abruptly: the
>> tracks would need to bend quite sharply to get into the corpus
>> callosum. You might want to relax the curvature constraint if that's
>> want you're trying to get.
>>
>> I don't have that many answers to this. Maybe there was a mistake in
>> the tracking? The FODs look OK on that coronal, but they do look odd
>> on the axial (not sure if I'm getting the slice orientations correct,
>> I'm used to human brains...). Are you using the correct file for the
>> tracking? Maybe you can copy/paste your entire command-line session to
>> see I can spot a problem somewhere...?
>>
>> Another problem could be some non-uniformity in the gradient
>> directions you use in your acquisition. That might cause strangely
>> consistent directions to appear that would otherwise be random noise.
>> Could you post your gradient encoding file? It's a text file with 4
>> columns in it, it might be called 'encoding.b'. If you don't have it,
>> that information might be stored in your dwi.mif file, you can extract
>> it with this command: "mrinfo dwi.mif -grad encoding.b".
>>
>> Cheers,
>>
>> Donald.
>>
>>
>> On Thu, Apr 9, 2009 at 7:25 PM, Wim Otte <wim at invivonmr.uu.nl> wrote:
>>> Hi Donald,
>>>
>>> Converting the reference.nii does the job! Thanks.
>>>
>>> Before running CSD on all gray-matter ROIS, I did some testing of the
>>> streamtrack application.
>>>
>>> What I notice is some straight line of high probability voxels through
>>> the brain (from one hemisphere to the other; without running through
>>> an anatomical struture). I know that false positives are still a
>>> problem to be solved in fiber tracking (review D. Jones), but I just
>>> want to ask if other people (using human data for example) encounter
>>> the same thing.
>>>
>>> I attached some screenshots from mrview (fODFs with ROI) and fslview
>>> (fa with tracks2prob result 1% - 100%).
>>>
>>> Is it normal to get these cross-like connectivities using prob. fiber tracking?
>>>
>>> Wim Otte
>>>
>>>
>>>
>>> On Wed, Apr 8, 2009 at 3:11 AM, Donald Tournier <d.tournier at brain.org.au> wrote:
>>>> Hi Wim,
>>>>
>>>> Yes, in theory, tracks2prob should just copy the layout from the
>>>> reference image. But it so happens that the NIfTI image handling
>>>> routine overrides the layout and sets it to +0,+1,+2. There is no
>>>> point in trying to use mrconvert to change the layout for a NIfTI
>>>> image, it will always end up with the same result. What you could do
>>>> is mrconvert the reference.nii image, and then the layouts will both
>>>> be +0,+1,+2. Does that sound like a workable solution?
>>>>
>>>> I might try to make the NIfTI handler honour the layout specification,
>>>> but that won't be ready for some time...
>>>>
>>>> Regards,
>>>>
>>>> Donald.
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Mrtrix-discussion mailing list
>>> Mrtrix-discussion at www.nitrc.org
>>> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>>>
>>>
>>
>>
>>
>> --
>> Jacques-Donald Tournier (PhD)
>> Brain Research Institute, Melbourne, Australia
>> Tel: +61 (0)3 9496 4078
>>
>



-- 
Jacques-Donald Tournier (PhD)
Brain Research Institute, Melbourne, Australia
Tel: +61 (0)3 9496 4078