The ISMRM 2015 Tractography Challenge

An example of processed data

Usage

We offer here examples of tractograms created on carefully preprocessed data. The preprocessing choices were made after the publication of the 2015 paper, with knowledge available from the analysis of the results presented above. Feel free to use. For instance, this could be used as training set in a machine learning experiment, to avoid training on the ground truth data. Please visit the References page to cite us if you use this data.

Data creation

Click to show details

The general steps are as in the description in the Tractoflow documentation. Data was run with an adapted tractoflow based on version 2.3.0.

  • DWI: Brain extraction, DWI denoising, Eddy/Topup, N4, Cropping, Normalization, Resampling, Extraction of fODF shells, Computation of fODF metrics.
  • T1: Denoising, N4, Resampling, Brain extraction, Croping, Registration on the DWI, Segmentation of tracking/seeding maps.
  • Tractography: seeding from the WM mask, npv 10, we offer here both local tracking and PFT tracking.

Modified steps were:

NoArtefact
  • DWI skull-stripping: Visually, we could see that there was no skull and little noise in the background. A simple threshold at value 50 allows a good brain extraction, rather than using BET.
  • DWI denoising, Topup and Eddy and N4 were skipped as DWI data was not noisy.
  • T1 Denoising and N4 were skipped.
  • T1 skull-stripping: Visually, we could see that there was no skull and little noise in the background. A threshold at value 50 allows a good brain extraction but creates some holes. The mask was dilated (4-pass) and eroded (4-pass) to fill the holes.
  • T1 and DWI were already well aligned and registration was skipped (affines were fixed to have the same affine for both files).
  • WM mask: Due to the fact that this is simulated data, dilating the white matter mask (2-pass) before using it as tracking/seeding mask allowed better results.
Basic
  • Topup was skipped, ignoring the reverse-phase b0.
  • T1 N4 step was skipped. Denoising was performed but N4 seemed to produce weird results on this simulated data. It was skipped.
WithReverse Same as above, except 2nd timepoint of the DWI was extracted, named rev_b0 and added to the input data to allow using the topup step.

Download

No artefact

Basic

With reverse-phase b0

Here are the results for these files.

Fig 4. WRONG TABLE FOR NOW. PREPARING STUFF. TODO.