[Mrtrix-discussion] mrtrix connectome

[Robert Smith]

Hi Basile (and all),

The difference between what was presented in the journal article
<http://www.sciencedirect.com/science/article/pii/S1053811912005824> and
the current ACT implementation is minuscule, and should only affect
tracking results in voxels with more than two tissue types (which are rare).

In terms of the image format, the difference is the inclusion of a fifth
tissue type (hence "5TT
<https://github.com/jdtournier/mrtrix3/wiki/Anatomically-Constrained-Tractography-(ACT)>"):
'pathological', which is the fifth of the five image volumes in the ACT
image. This allows users to manually delineate regions of the brain where
the nature of the underlying biology is unknown; streamlines in such
regions are allowed to propagate or terminate freely, i.e. without the
application of any priors from the anatomical image.

I definitely agree that tissue segmentations that provide partial volume
images rather than binary masks should give better tracking results. If you
think of the GM-WM interface as a surface, by converting your tissue
estimations into binary masks you are immediately introducing jagged edges
into your reconstruction of that surface; whereas with good partial volume
estimates, the sub-voxel surface along which the interpolated GM and WM
fractions are equal will approximately follow the actual underlying surface.
However this is not the only issue that I have with using FreeSurfer tissue
segmentations for ACT: FreeSurfer's strength is delineation of the cortical
ribbon, but it fails to segment many mid-brain structures, so the tracking
does not behave as expected in those areas. If anything, this script was
provided to demonstrate how tissue segmentations from any alternative
software package can be manipulated into a format appropriate for ACT.

The console message "*resampling ACT 5TT image to fixel image space"* is
appearing due to the use of dynamic seeding. This seeding method constructs
the same model for comparing FOD amplitudes to streamlines densities as is
used in the SIFT method
<http://www.sciencedirect.com/science/article/pii/S1053811912011615>, and
therefore requires the same definition of the processing mask in diffusion
image space as is used in that method; the dynamic seeding itself is
however unpublished. The precise calculation of this mask also differs
slightly from that described in the ACT paper, but this response is long
enough as it is...

Regardless of whether or not you are using dynamic seeding, *the tracking
itself always uses the ACT 5TT image at its provided resolution*. This is a
big benefit of performing tracking in 'real space' as is done in MRtrix:
there's no need to explicitly re-sample one image to match another and
hence degrade the image resolution. From a streamline's position in real
space, a single transformation and interpolation step is performed to
access the diffusion image information, and (independently) another
transformation and interpolation step is performed to access the tissue
information. You should therefore provide the tissue segmentation image at
the resolution at which it is generated.

Best regards
Rob

--

*Robert Smith, Ph.D*
Research Officer, Imaging Division

The Florey Institute of Neuroscience and Mental Health
Melbourne Brain Centre - Austin Campus
245 Burgundy Street
Heidelberg Vic 3084
Ph: +61 3 9035 7128
Fax: +61 3 9035 7301
www.florey.edu.au


On Sat, Jun 21, 2014 at 12:57 AM, basile pinsard <basile.pinsard at gmail.com>
wrote:

>   Hi mrtrix experts and users,
>
>  could you provide me some details about ACT in practice?
>  is the current version method different from what is described in the
> article?
>
> Partial volume (pv) seems to give better results than converted freesurfer
> aparc binary maps but tckgen says:* resampling ACT 5TT image to fixel
> image space*
>  is that only to create the seeding mask used for dynamic seeding? in
> this case is that just downsampling white matter pv to the scan resolution
> and thresholding at .5? or to the tracking step resolution? or else?
>  Is this resampling also performed for the partial volume values or does
> it interpolates dynamically from native provided 5TT?
> What is the best 5TT image to provide the software, is the highest
> resolution achievable useless?
>
>  Many thanks.
>
>  Cheers
>
> --
>  Basile Pinsard
>
> *PhD candidate *
> Laboratoire d'Imagerie Biomédicale, UMR S 1146 / UMR 7371, Sorbonne
> Universités, UPMC, INSERM, CNRS
> Unité de Neuroimagerie Fonctionnelle, CRIUGM, Université de Montréal
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20140621/cd807acd/attachment.html>