Diffusion - McIntosh-Lab/tvb-ukbb GitHub Wiki

Processing steps for diffusion imaging data that we have retained from the UK Biobank pipeline include correction of eddy currents and head motion (EDDY), diffusion tensor image fitting (DTIFIT) for tract-based analysis (TBSS), and multi-fibre orientation modelling (BEDPOSTX) (Figure 4). New features and additions to the diffusion sub-pipeline are described below.

_{Figure 4: Diffusion sub-pipeline workflow. Original components of the UKBB pipeline with few or no modifications shown in purple; pipeline components with major changes or additions shown in white; and new components shown in orange. Dotted lines indicate components that are included in the QC report. Black lines indicate components that are included in the TVB Inputs.}

Our first addition to the diffusion sub-pipeline was the integration of B0 field estimation for unwarping data that lack reverse phase-encoded images using the Synb0-DisCo tool (Schilling et al., 2019). This tool uses a deep learning approach to create a synthetic undistorted B0 image from a T1w image. The synthetic undistorted B0 is used as input to FSL’s TOPUP toolbox for dMRI distortion correction. In our pipeline, users have the option to implement this tool to improve registrations between the T1w and dMRI images.

The other major addition to the dMRI sub-pipeline was the replacement of the UK Biobank tractography approach with one that takes as input the user-defined parcellation for connectome construction. In our approach, the grey matter–white matter labelled interface is registered to the distortion-corrected B0 image. This interface is used to define seed and target ROI masks. The grey matter mask is also registered to the B0 image and used as an exclusion mask. Probabilistic tractography is performed using FSL’s PROBTRACKX toolbox to generate a matrix of the streamlines between all ROIs. The structural connectivity ‘weights’ matrix is then computed by taking the streamlines matrix and dividing it by the total number of streamlines that were successfully sent from the seed ROIs. This weights matrix therefore encapsulates the probability of connection between all ROIs. ‘Distance’ matrices (i.e., estimated tract lengths) are also retained.