Tract-based spatial statistics (TBSS) Preprocessing

Tract-based spatial statistics \cite{Smith_2006} is a method of white matter analysis applied to diffusion-weighted images (DWI). DWI is a measure of water diffusion in the brain and is generally considered an indicator of fiber tract integrity. DWI measures include fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD). FA values quantify how strongly directional local tract structure is, thus FA measures are sensitive to microstructural changes in the WM. The FA values range from 0 to 1, with white matter (WM) having values of 1, wherein diffusion is anisotropic, and CSF having values of zero. MD is total movement of water in a voxel. It is an indicator of cellular integrity, as it increases with cellular damage. AD is a measure of diffusion along the strongest direction, whereas RD is a measure of diffusion perpendicular to that. TBSS is an automated methods that combines the strengths of voxel-based morphometry (VBM) and tract-based spatial statistic methods. As in VBM analyses, participants FA image is registered to a template so that differences between groups can be determined using voxelwise statistics. However, because coregistration algorithms are not able to accurately align long white matter fiber tracts that vary in size and shape across participants, TBSS relies on registering participants to a common template space, providing a bit of an improvement over traditional VBM. 
There are three preprocessing steps for TBSS . 1) Use linear registration tools in the FMRIB Software Library (FSL) eddy_correct program (downloadable at https://fsl.fmrib.ox.ac.uk/fsldownloads_registration) to register the DWI to the non-diffusion-weigthed image (the b0 image) for motion correction. The input image is the NifTi-formated DTI image, and the output is the motion corrected image. 2) Use the FSL brain extraction tool (BET) program \cite{Smith_2002} to skull strip and exclude non-brain voxels. The input image is the output image from the previous step. The output image is a brain mask. The -f option can be adjusted to improve the brain extraction process. The -m flag tells the program to save the mask as an ROI.  3) Calculate diffusion tensors using the FSL DTIfit program for whole brain volumes. The results of the final step are the FA, RD, MD, and AD volumes necessary for the tract-based spatial statistic analysis. The specific code for these three major steps follows as: