ISMRM 2015 Tractography challenge - Discussion

The last hour of the Diffusion Study Group of the ISMRM 2015 was dedicated to a discussion and debate on the open questions of tractography, using insights gained from the results of the ISMRM 2015 Tractography Challenge.

Some of the main insights that were presented include:

  • Generating streamlines that are perfectly voxel-aligned to the ground-truth anatomy is really challenging, in part because of the artefacts correction process. These corrections almost always add some shift with respect to the underlying anatomy.
  • The previous point can have a huge impact in a pipeline that registers multiple subjects to a common template, for example for group studies.
  • Scores are much better and more realistic when using a bundle recognition technique, instead of the classical Tractometer technique, which is based on endpoints.
  • Most teams found 23 of 25 Valid Bundles, at the cost of generating multiple Invalid Bundles.
  • Only one submission found the 25 Valid Bundles.
  • The most difficult bundles were the Anterior Commissure and the Posterior Commissure.
  • The main differences between pipelines are seen in the Invalid connections and the "No connections" (streamlines not connecting 2 regions).
  • There needs to be a unified tractography format, to avoid potential interaction and interpretation problems.
For more insights, you can go through the results presentation document.

UPDATE: The results of the analysis have been accepted for publication. See the paper page.