Automatic white matter segmentation

We propose an automatic method to classify the brain fibers generated by the tractography of a subject.

  • The project was carried out on data provided by the O'Donnell Research Group from the Human Connectome Project (HCP)[1]. The objective is to classify the fibers into 57 white matter areas called tracts.
  • To classify a 3D shape we use here a method based on multi-view representation. We used the fly-by-cnn project[2] to build a sequence of images representing each fiber at different locations in space.
  • Then, we used an RNN to predict the class of each fiber from this sequence as well as its confidence in the prediction made.
  • Next, we used the MRtrix3 library [3] to apply the iFOD2 algorithm and create a second, less clean test dataset to test the accuracy of our network.

[1] https://github.com/SlicerDMRI/ORG-Atlases

[2] https://github.com/DCBIA-OrthoLab/fly-by-cnn

[3] https://www.mrtrix.org