Pipeline for DTI processing

Diffusion weighted images have to be preprocessed before any analysis. The DTI processing pipeline include the following steps:

  • non-uniformity correction of the anatomical
  • registering multiple DWI sets
  • registering all diffusion images to the anatomical
  • creating a brain mask
  • calculating eigenvectors and eigenvalues of diffusion tensor

Registration of DWI sets

As a general rule, all diffusion weighted images should be registered to each other. However, because the contrast between images acquired in different directions can be significantly different, conventional registration approaches might fail. In the meantime, we often assume the subject has not moved within a run and we only register the b=0 image to the anatomical.

Averaging of DWI sets

In the case where multiple diffusion sets were acquired, increased SNR can be achieved by combining the sets. However, care must be taken in doing so. Some assumptions have to be made at this point, which are probably inconsistent with reality but seems to be the current approach in the literature. First, we assume that there is no motion across a set. Second, we assume that if there was motion, it occured during the time between the acquisition of the sets (i.e. DW set 1 in position 1, DW set 2 in position 2).

Creating a brain mask

Calculating eigenvectors and eigenvalues of diffusion tensor