[Mrtrix-discussion] On trials parameter in streamtrack

[J-Donald Tournier]

Hi Alessandra


> I was wondering how streamtrack work in generating and selecting tracks.
> As far as I could see from mailing list, number of streamlines generated
> are usually higher than actually selected. Both commands can be set by
> -maxnum and -number options.
> Reasons for excluding a streamline depend basically on tract length,
> inclusion or exclusion rois used.
> To get somehow a sense of the probability to get a connection one could
> divide number of selected streamlines to the number of generated ones.
>

Yes, that is one option, but I'd stress that you really want to avoid any
interpretation of that number as a measure of actual probability. This is
the reason why Geoff Parker talks about a probabilistic index of
connectivity (PICo) <http://www.ncbi.nlm.nih.gov/pubmed/12884338> rather
than actual probabilities - there are far too many factors that come into
it, see for example Derek Jones' excellent paper on the topic
<http://www.ncbi.nlm.nih.gov/pubmed/22846632>.


> Looking at the messages in the archives, during probabilistick tracking,
> seed points are randomly selected within the seed mask, then a direction is
> chosen and eventually stream process starts if amplitude in that direction
> is higher than initcutoff.
> From my point of view an approach where a certain number of attempts are
> made from each voxel in brain mask would make a lot of sense to provide
> meaningful probability results related to "existence" of a connection;
> giving the opportunity to choice output number could be potentially
> dangerous, expecially when making connectomics analyses.
>

Well, I'd disagree to some extent with the word 'meaningful' in this
context, but yes, this is a common request. This option is actually
available in MRtrix3 (along with lots of other approaches to seeding).
Unfortunately, we have no plans to back-port that feature to the MRtrix 0.2
version.


> I see there exists an option in streamtrack, -trials. How it does work
> precisely?
> I tried with a single voxel seed, fixing maxnum and number (100 or 500)
> and varying trials parameter. Considering i have set a negligible cutoff
> for minlength (1mm), what i see is that results strongly depend on trials
> parameter. Really small trials (5) cause all tracts terminating quite close
> to seed point; with higher number of trials, results get better. On the
> other side, the larger this number, the higher the likely to get spurious
> connections.
> Is it correct to say, due to request of a certain number of streamlines, a
> lot of seed points are initializied within that unique seed voxel, and only
> after that trial parameter gets effective?
>
I am asking because I am not interested in making whole brain analysis,
> hence I want to carefully inspect how these parameters work. In the light
> of that, what could be a good compromise when handling trials, number and
> maxnum parameters?
>

The -trials option is unrelated to what you're interested in - it's to do
with the rejection sampling used during probabilistic tracking, and really
shouldn't need to be adjusted. What you want in this scenario is to set
-maxnum and -number to the same value.

>
I would say if you need finer control over the seeding strategy, you'll
want to try MRtrix3 <https://github.com/MRtrix3/mrtrix3> instead. And if
you're seriously interested in connectomics, I recommend you think about
including ACT <http://www.ncbi.nlm.nih.gov/pubmed/22705374> & SIFT
<http://www.ncbi.nlm.nih.gov/pubmed/23238430> to generate connectomes with
meaningful track counts - both of which are available in MRtrix3 and
documented in their respective wiki entries (ACT
<https://github.com/MRtrix3/mrtrix3/wiki/Anatomically-Constrained-Tractography-%28ACT%29>,
SIFT <https://github.com/MRtrix3/mrtrix3/wiki/SIFT>), along with our
recommendations
regarding connectomic analysis
<https://github.com/MRtrix3/mrtrix3/wiki/Structural-connectome-construction>
.

Hope that helps,
Donald.



>
> Best,
> Alessandro
>
> --
> Alessandro Calamuneri, M.Sc., PhD
> Department of Neurosciences, University of Messina, Italy
>
>
> _______________________________________________
> Mrtrix-discussion mailing list
> 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 EngineeringKing's College London*


*A: Department of Perinatal Imaging & Health, 1st 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
<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/20150617/748db2a0/attachment.html>