Tractography gold standard pipeline

Dear All,

I am new to mrtrix. I have two 20 subejct DWI sets (60 directions, 2^3 mm^2 voxels, b value set =700 1st set, b=3000 2nd set). Given the large array of tractography and anatomically constrained routines in mrtrix, we need to do a (mostly internal) parameter space search study to optimize processing against a ‘gold standard’. Could you recommend a pipeline that:

  1. provides as “realistic” gold standard (close to in vivo brain human data)
  2. readily accessible/publicly available
  3. Readily available goodness-of-fit measures.
    We basically look for a pipeline to perform our parameter study that will be efficient time-wise but having certain generalizability. Rigor is important; rigor aside, time is of the essence.

Please advise, thank you


Hi Octavian,

I think it may help to disambiguate precisely what you are looking for in terms of a “gold standard”:

  • If you are using a reconstruction pipeline to define a “gold standard”, then any “goodness-of-fit measure” will only be capable of comparing any other reconstruction to that “gold standard”, which itself isn’t necessarily correct (since it is itself generated from tractography). Moreover, by definition the pipeline you used to define the “gold standard” will always provide the best fit, likely rendering a parameter search redundant.

  • If you want a more robust “gold standard” to which goodness-of-fit measures are available and a reasonable indicator of performance, you probably need to be looking into DWI phantoms. A reasonable amount of parameter searching has already been performed using Tractometer as an example.

Hope that helps you in finding a direction for your research

Thank you, Rob
re Tractometer. Can one share the most successful pipelines using mrtrix tools used in ISMRM 2015 challenge or related studies such as
Neher PF, Descoteaux M, Houde JC, Stieltjes B, Maier-Hein KH. Strengths and weaknesses of state of the art fiber tractography pipelines–A comprehensive in-vivo and phantom evaluation study using Tractometer. Med Image Anal. 2015 Dec;26(1):287-305.
I fully understand that generally pipelines have to be adapted to particular raw data and scope under study, and also suffer timely updates given rapid mrtrix upgrades; mrtrix offers that versatility to build these. Still, it would nevertheless be nice to have generic pipeline full codes available and amenable for tweaking. For example, we are interested in structural (global) connectome building using two 60 dir, opposite AP/PA phase encode direction sets using single shell data. Is there a generic ‘full’ pipeline code available from the mrtrix authors,
thank you.

The MRtrix3 developers mostly haven’t participated in such competitions, largely because we could foresee from the data / quantitative metrics that our methods would likely be rated poorly, though this would not be reflective of the performance of our methods on real data and would be more a consequence of the data / metrics used. I had hoped to participate in the ISMRM 2017 tractography challenge, but simply ran out of time; we’ll see if they make the data / automated assessment available in an ongoing fashion.

Alternatively, there is the MRtrix3_connectome script I wrote as part of the Stanford Centre for Reproducible Neuroscience coding sprint. It’s out of date and won’t work against MRtrix3 version 3.0_RC1 and requires further developments and isn’t considered “a part of MRtrix3”, but it might serve as a useful reference / starting point for you.