[Mrtrix-discussion] RE: Mrtrix tractography questions

Donald Tournier d.tournier at brain.org.au
Thu Jun 16 18:49:40 PDT 2011


Hi Marc,

We have a couple of questions regarding the tractography mechanics involved
> in MRtrix
>
> 1. How does Mrtrix start sampling tracts from ROI’s? That is, if you
> provide an ROI how does MRtric decide where to start the tract from? From
> the boundary of the ROI or from the centre? Do you take into account if the
> ROI overlaps grey and white matter?
>

The seeding is random, with uniform probability over the whole ROI. For a
spherical ROI, any point within the specified radius of the specified centre
is a candidate. For a mask ROI, any point with intensity greater than 0.5
after tri-linear interpolation is a candidate.

In terms of dealing with the overlap with grey/white matter: basically it'll
try to seed regardless, but only succeed if the seed point and direction are
appropriate. First a seed point is drawn from the ROI. Then a starting
orientation is drawn at random with uniform probability over the sphere
(unless initdir has been specified). Finally, if the starting conditions are
satisfied, tracking will start from that seed along that direction. For
tensor tracking, the starting condition is simply that the FA is above
initcutoff. For SD-based tracking, the starting condition is that the FOD
amplitude along the starting direction is above initcutoff.



> 2. We are also interested in your thoughts on putting a “weighting” on
> tract length.
> A number of studies suggest that if were using a simple deterministic
> tracking algorithm to measure connections between multiple ROI’s we should
> weigh long tracts since there is bias towards longer tracts. Does the same
> apply for your algorithms? We were even suspecting that using the
> probabilistic tracking option there would be bias against longer tracts?
>

That's the first I've heard about penalising long tracts for deterministic
tracking - maybe that's just a reflection of the fact that I'm somewhat
biased towards probabilistic approaches... In any case, for probabilistic
tracking (including streamtrack's SD_PROB method), the consensus is that
there is a bias towards shorter tracks. How best to deal with it is
something that still hasn't been addressed satisfactorily. Probably the most
relevant work in this area is Morris et al., Neuroimage 42: 1329-1339. I'm
not convinced this will necessarily solve the bigger problem though, which
is how to properly interpret these 'connectivity' measures, a problem that
Derek Jones describes very eloquently in his 2010 paper "Challenges and
limitations of quantifying brain connectivity in vivo with diffusion MRI",
Imaging in Medicine 2: 341-355.

Cheers,

Donald.




>
>
>
> cheers
>
> Marc
>
>
>
> *Marc Seal
> **Group Leader, Developmental Imaging, Murdoch Childrens Research
> Institute
> Senior Research Fellow, Department of Paediatrics, The University of
> Melbourne
>
> *The Royal Children's Hospital, Flemington Road, Parkville, Victoria 3052,
>  Australia
> M  0423 606 185  |  T +61 3 9345 4310   |  F   +61 3 9345 4325
> E   marc.seal at mcri.edu.au
> www.mcri.edu.au
>
>  Please consider the environment before printing this email
>
>
>
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email
> ______________________________________________________________________
>
> _______________________________________________
> 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 9035 7033
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20110617/b26948e7/attachment.html


More information about the Mrtrix-discussion mailing list