I learned from these two sources:

https://andysbrainbook.readthedocs.io/en/latest/MRtrix/MRtrix_Introduction.html

https://micapipe.readthedocs.io/en/latest/pages/02.dwiproc/index.html

Required Softwares:

  • MRtrix3 (You can find how to download here)
  • FSL
  • ANTS (Installation instructions here)
  • FreeSurfer (if you need surface base parcellation, download here)

DWI Processing

Required Documents: 3 types of files .nii, .bval, .bvec.

  • .nii contains the brain image
  • .bval indicates the b value of each volume
  • .bvec indicates the b vector

Usually, we need to specify the predominant phase-encoding direction (e.g., Anterior to Posterior, or AP) and the reverse phase-encoding direction (e.g., PA) and extract the two sets of images.

From Preprocessing to fiber tractography streamline, we need to go through the following steps:

  • Denoising:
    • MP-PCA denoising
    • A reverse phase encoding
    • Optional: Correction of susceptibility distortion (e.g., Gibbs’ ringing artifacts),
    • Eddy current-induced distortions and motion,
    • Non-uniformity bias field correction.

Then DTI scalars such as the Fractional Anisotropy (FA) and Mean Diffusivity (MD) can be computed.

  • Basis function for each tissue type
    • The Dhollander algorithm is used to estimate the response functions of cerebrospinal fluid (CSF), gray, and white matter.
    • These are then used to estimate the fiber orientation distribution (FOD) by spherical deconvolution.
    • Intensity normalization is applied to each tissue FOD to enable comparison between subjects.
  • Create a GM/WM boundary for seed analysis
    • A non-linear transformation is computed between the normalized white matter FOD and s-MRI (T1) aligned to the b0 image.
    • Then, a five-tissue-type (5TT) image segmentation is generated and registered to the DWI space, and a gray matter white matter interface mask is calculated.
  • Run the streamline analysis
    • First, a tractography with 10 million streamlines is generated using the iFOD2 algorithm
    • Next, spherical deconvolution informed filtering of tractograms [SIFT2] is applied to reconstruct whole brain streamlines weighted by cross-sectional multipliers.