[Mrtrix-discussion] question about CSD

[Donald Tournier]

Hi Michael,

>From my point of view, the reason for it is purely a matter of
resolution and the corresponding partial volume effects. If your fibre
bundle is oriented through-plane, and assuming that it is well-behaved
(i.e. it curves relatively smoothly with no sudden kinks, etc.), then
its cross-section is well-characterised using the acquisition you
describe, and partial volume effects will not be as big a confound.
The tracks will run along the direction of low resolution, and the
tracking algorithm will be able to make use of the increased spatial
resolution in the other directions to generate good tracks.

If your bundle was aligned in-plane though, its upper and lower edges
(assuming an axial acquisition) will suffer from severe partial volume
effects, which may confound the tracking. I would expect this might
introduce some bias between the in-plane and the through-plane case,
even if that translate only into a simple blurring of the in-plane
tracks along the through-plane direction.

Not sure I'm making sense, or whether what I'm saying is in fact
correct. Like I said, I haven't looked into this effect at all, and
I've always used acquisitions with isotropic voxels in order to avoid
having to deal with precisely that issue...

Cheers,

Donald.


2010/1/19 Michael Zeineh <mmzeineh at gmail.com>:
> The reason you think it would be better through plane is what ... Better SNR
> (I would think SNR just depends on overall voxel size)? Less eigenvector
> error for some other reason?
>
> Let me pose it more specifically. Say our in-plane resolution is say 1mm,
> and our thru plane resolution is 5mm. Let us say the white matter structure
> has detail along two but not three of its dimensions that could not be
> resolved at 5mm resolution but could be resolved at 1mm resolution. Along
> the third dimension, it could be resolved at lower resolution. In this case,
> one would definitely want the orientation such that the two dimensions
> requiring the resolution are the in-plane orientation with higher
> resolution, I would think.
>
> Olivier, it would be great to get the insight from the talk if this issue
> was addressed.
>
> -----Original Message-----
> From: Donald Tournier [mailto:d.tournier at brain.org.au]
> Sent: Sunday, January 17, 2010 8:30 PM
> To: Michael Zeineh
> Cc: mrtrix-discussion at www.nitrc.org
> Subject: Re: [Mrtrix-discussion] question about CSD
>
>
> Hi Michael,
>
> Actually, I would have thought it would be the opposite: a structure
> entirely through plane would be resolved better. That said, I can't think of
> any studies that have looked at that specifically, so don't quote me on
> that... Maybe someone else on the list can suggest an appropriate
> reference...?
>
> Cheers,
>
> Donald.
>
>
> 2010/1/18 Michael Zeineh <mmzeineh at gmail.com>:
>> Thank you Donald.
>>
>> I see. So, for the example an axial DTI with thick slices but small in
>> plane voxels, a structure entirely in-plane would be resolved better
>> than if the same object were entirely through-plane (assuming things
>> like SNR are similar).
>>
>> Michael
>>
>> On Sun, Jan 17, 2010 at 3:18 PM, Donald Tournier
>> <d.tournier at brain.org.au> wrote:
>>> Hi Michael,
>>>
>>> Yes, it should work with anisotropic voxels (although isotropic would
>>> always be recommended). There is little point interpolating (at least
>>> not using linear interpolation), since the tracking code performs
>>> linear interpolation while tracking. In terms of bias, the
>>> orientations are provided with respect to real (scanner) coordinates,
>>> so do not depend on the voxel dimensions. There would however be a
>>> bias when tracking WM structures oriented predominantly through-plane
>>> versus in-plane, since the "effective resolution" would be higher in
>>> the first case. This applies to all tracking methods though, not just
>>> MRtrix (and is not a limitation of the CSD itself).
>>>
>>> Hope that helps,
>>>
>>> Donald.
>>>
>>>
>>> 2010/1/16 Michael Zeineh <mmzeineh at gmail.com>:
>>>> Out of curiosity, will it work on anisotropic diffusion data (i.e.
>>>> voxels are thicker along one axis)? If so, would there be any
>>>> expected errors or biases? Would simple interpolation (somewhat)
>>>> resolve those issues?
>>>>
>>>> Thank you,
>>>> Michael
>>>> _______________________________________________
>>>> 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
>
>



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