DTI analysis: An outline

An outline of DTI dataset processing pipeline for TBSS analysis as well as probabilistic tractography using FSL.

DTI data processing

1. Data preparation
   1. Convert dicom files to nifti
      1. Use dcm2nii or dcm2niigui from mricron
   2. Rename files using the following scheme:
      1. dti.nii – for DTI images
      2. dti.bval – for the bval file
      3. dti.bvec – for the bvec file
      4. anat.nii – for T1 anatomical file
   3. create a text file listing the directories containing the DTI files, one directory per subject (e.g., dirlist.txt)

2. Preprocess DTI datasets
   1. Script: fn_dtipreproc.sh [dirlist] [TBSSdir]
   2. [dirlist] – text file containing directory list (see 1.c)
   3. [TBSSdir] – optional, directory where output will be saved, default value is tbss in the current directory, will be created by the script if doesn’t exist
   4. The script performs the following operation
      1. Correct for effects of movement and eddy current by running eddy_current
         1. Output is dti_ecc.nii.gz
         2. Assumed first volume in dti.nii is the B0 volume
      2. Extract B0 volume from dti.nii using fslroi
         1. Output is dti_nodif.nii.gz
         2. Assumed B0 is first volume in dti.nii
      3. Create brain mask using bet and B0 image
         1. Output is dti_nodif_brain.nii.gz – extracted brain (betted), dti_nodif_brain_mask.nii.gz – binarized extracted brain
      4. Fit the diffusion tensor model to the data using dtifit and the dti_ecc image
         1. Output files has dti prefix and stored in the same location as the original image
            • dti_V1, _V2, _V3 – 1^st, 2^nd, and 3^rd eigenvectors
            • dti_L1, _L2, _L3 – 1^st, 2^nd, and 3^rd eigenvalues
            • dti_MD – mean diffusivity
            • dti_FA – fractional anisotrophy
            • dti_MO – mode of anisotrophy (oblate: ~-1, isotrophic: ~0, prolate: ~1)
            • dti_SO – raw T2 signal without diffusion weighting
      5. Copy dti_FA.nii.gz to the TBSS directory (see 2.c) and renamed it as <subjdir>_FA.nii.gz, where <subjdir> is the subject directory
      6. After processing all subjects’ data, the script runs the following commands in the TBSS directory
         1. tbss_1_preproc *.nii.gz
            • prepare FA data in the right format
         2. tbss_2_reg -T
            • apply nonlinear registration of all FA images into 1x1x1 standard space
            • -T : target is FMRIB58_FA
         3. tbss_3_postreg -S
            • create mean FA and skeletonized it
            • -S : use derived mean from subjects’ data
         4. tbss_4_prestats 0.2
            • project all subjects’ FA onto the mean FA skeleton
   5. references
      1. http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/tbss/UserGuide
      2. http://wikis.la.utexas.edu/imagelab/book/how-prepare-your-dti-images-fractional-anisotropy-fa-analysis

3. TBSS analysis (tract-based spatial statistics)
   1. Do two-sample t-test using randomise
      1. Change directory to [TBSSdir]/stats

          [user]$ cd [TBSSdir]/stats

      2. Generate design matrix and contrast using design_ttest2 for simple model

          [user]$ design_ttest2 OUTPUTNAME N1 N2

         1. This will output OUTPUTNAME.mat and OUTPUTNAME.con, which you will need in randomise
         2. N1 is the number of participants in group 1
         3. N2 is the number of participants in group 2
         4. Make sure that you have the correct order by running imglob to check what group is listed first

             [user]$ imglob ../origdata/*.nii.gz

      3. Run randomise

          [user]$ randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d OUTPUTNAME.mat -t OUTPUTNAME.con -n 5000 --T2

         1. this will output several files with “tbss_” prefix
         2. of interest is the file named tbss_tfce_corrp_tstat*.nii.gz, which are the TFCE corrected p-value image
      4. Check results

4. Run bedpostx – fitting probabilistic diffusion model on corrected data
   1. Script: fn_bedpostx.sh <dirlist>
   2. <dirlist> – a text file containing directory list (see 1.c)
   3. This script performs the following operation
      1. Create tracto/ directory in the subject directory as output location
      2. Copies relevant files from subject directory to tracto directory
         1. nii.gz to data.nii.gz
         2. bval to bvals
         3. bvec to bvecs
         4. nii.gz to nodif_brain.nii.gz
         5. nii.gz to nodif_brain_mask.nii.gz
      3. Creates condor job description and save it in <dirlist>_bedpostx.job
      4. Submits job to condor for processing
      5. Output will be in the subject directory tracto.bedpostX

5. Register DTI images to standard MNI space/template
   1. Script: fn_dtiregister.sh <dirlist>
   2. <dirlist> – a text file containing directory list (see 1.c)
   3. Output will be in the bedpost directory, i.e., tracto.bedpostX/xfms
      1. Linear
         • DTI -> structural: diff2str.mat
         • Structural -> DTI: str2diff.mat
         • Structural -> standard: str2standard.mat
         • Standard -> structural: standard2str.mat
         • DTI -> standard: diff2standard.mat
         • Standard -> DTI: standard2diff.mat
      2. Nonlinear
         • Structural -> standard: str2standard_warp.nii.gz
         • Standard -> structural: standard2str_warp.nii.gz
         • DTI -> standard: diff2standard_warp.nii.gz
         • Standard -> DTI: standard2diff_warp.nii.gz
   4. NOTE: be sure to check ranat_brain.nii.gz if bet is working properly. If not, you may need to run bet manually.

After step 5, you’re ready to run probabilistic tractography. Just run Fdt and enter the needed parameters in the GUI.