DTI Brain Network Construction

The DTI Brain Network Construction pipeline is a standardized pipeline for general researchers to build their own structural brain network from diffusion tensor imaging (DTI) MRI data.
The pipeline uses FSL to perform eddy and motion correction. The Diffusion Toolkit from TrackVis is used to reconstruct local diffusion patterns and tractography. The brain region parcellation was completed with both FSL and Freesurfer. The Matlab is used to adjust bvecs for local diffusion pattern reconstruction and network computation.
All of those scripts already packed the command of the tools we used inside of them, allowing for minimum prior knowledge required from users to construct their own network.


System requirement:

  • Linux or MacOS (Intel Core)


Expected Data To Begin

  • DTI MRI raw data, for example, named 3101.nii.gz
  • bvec and bval gained from sMRI data, for example, named 3101_r.bvec and 3101_r.bval

Step 1: Preprocessing with FSL

Raw imaging data usually come with eddy current distortion and motion distortion. Therefore, we could use FSL eddy_correct to preprocess the data. Also, for better analysis result, we need to remove unnecessary brain tissue such as bones, dura, air, etc. by creating brain mask.

  • Preprocess the DTI data
    eg: 3101.nii.gz
    # command format: eddy_correct input_image_path output_filename_path index
    # output files: {name}_EC.ecclog, {name}_EC.nii
    eddy_correct 3101.nii.gz 3101_EC 0
    # command format: bet input_image_path output_filename_path -m -n -f 0.2
    # output files: {name}_brain_mask.nii.gz
    bet 3101.nii.gz 3101_brain -m -n -f 0.2

Step 2: Correct gradient table and erode mask

Since we are using FSL for eddy correct and motion correction, we should create a unique gradient table for every subject. We need to adjust it with the ecclog produced when converting dicom file to nii, and rotate it for further usage.

  • Process with Matlab code adjust_bvecs.m
  • Use the processed DTI's corresponding structural MRI's bvec and bval
    eg: 3101_r.bvec, 3101_r.bval, 3101_EC.ecclog (from Step 1)
    % command format: adjust_bevecs(.ecclog from Step 1, .bvec prepared, output adjusted {name}.bevc)

For better brain tractography result, the high value points at the brian edge should be removed using the erosion.

  • Process with Matlab code myerode.m
    eg: 3101_brain_mask.nii.gz (from Step 1)
    % addpath('path to downloaded NIfTI')
    % command format: myerode('path to brain mask generated in Step 1','output brain mask path')
    % output files: {name}_mask.nii

Step 3: Generate whole brain tractography

We need to run the brain tracktography algorithms to find the fiber tracts connecting different brain regions.

  • Edit the first 15 lines of the script run_dtk.sh to the following format
    • Diffusion Toolkit installed path
      export PATH="$PATH:/bin/dtk"
      export DSI_PATH="/bin/dtk/matrices"
    • FSL installed path
      export FSLDIR="/usr/local/fsl"
          . "${FSLDIR}"/etc/fslconf/fsl.sh
      data="(path to folder)/3101_EC.nii.gz"
      bvec="(path to folder)/3101_r.bvec"
      bval="(path to folder)/3101_r.bval"
      FA="(path to folder)/3101_mask.nii"
      temp="(path to folder)/temp"
      output="(path to folder)/dtk"
  • Then run the following code in bash. The generated brian tracktography graph would be placed as a folder in the subject folder named dtk.

Step 4: Label Generation

We need to determine region of interests (ROIs) and register them from the structure MRI to the brain tractography graph.

  • Change the following lines of the reg_FSROI_Jan2014.sh to the following format

    • line 7: Freesurfer installed path

      export FREESURFER_HOME="/usr/local/freesurfer/7.2.0"
    • line 21: FSL installed path

      export FSLDIR="/usr/local/fsl"
    • line 88 & 89 change the second arguments in these two line to the final output of step 2

      ${FSLDIR}/bin/flirt -in brain-in-rawavg.nii -ref {name}_mask.nii -out brain-in-DTI.nii.gz -omat TM.mat -cost corratio -dof 12  -interp nearestneighbour
      ${FSLDIR}/bin/flirt -in FS_label.nii.gz -ref {name}_mask.nii -out label.nii.gz -applyxfm -init TM.mat -interp nearestneighbour
  • Then run the following code in bash. The output would be placed in the subject folder name label.nii.


Step 5: Compute Network

Finally, we could get the brain network based on the registered brain tractography graph. The algorithm could calculate the structural connectivity based on the fiber tracks connecting two regions.

  • We need to first overlay the label.nii and dtk\dti_fact_tracts.trk in the TrackVis to make sure there is no orientation mismatch.
  • Run compute_matrix_Dec_11_2015.m
    % command format: compute_matrix_Dec_11_2015(`trk` file from step 3,`label.nii` from step 4, output path, threshold)
    % output files: .mat file containing constructed brain network martrix