DTI Notes | Home

Basic DTI terminology
Diffusion Tensor Imaging (DTI): a magnetic resonance imaging technique that uses single-shot echo-planar MRI pulse sequences which have been optimized to detect the diffusivity of water molecules in tissue. By inducing magnetic gradients in different non-colinear directions while a subject is in the scanner, a series of diffusion weighted images (DWIs) can be acquired. For each given gradient, the echo attenuation values (ie "intensities") at each voxel can then be used to calculate the apparent diffusion coefficient (ADC). The set of ADCs for each voxel can be used in turn to calculate the diffusion tensor for that voxel, an ellipsoid which models the diffusion profile of water within that voxel. The eccentricity of the ellipsoid is known as the Fractional Anisotropy (FA); the mean of the three principle eigenvectors is known as the mean diffusivity (MD).

Tractography: determination of the directional pattern of diffusion tensors. This is often represented as a color image in which green represents anterior-posterior, red represents left-right, and blue represents head-foot/dorsal-ventral. Methods including streamline tractography (eg FACT) or probabilistic tractography (eg BEDPOST) can be used to perform connectivity analyses.

Fractional Anisotropy (FA): a measure of the diffusivity at each voxel, expressed as a ratio between 0 and 1, where complete isotropic diffusivity is 0 and complete anisotropic diffusivity in a single direction is 1.

Gradient vector: a 3D set of coordinates representing the direction (vector) along which a magnetic gradient has been applied. Each DTI pulse sequence has a set of gradient vectors associated with it, which is required for subsequent analysis.

B-value: a scaling factor used to describe the attenuating effect of the MR signal on the different gradients.

B0 (aka "B not"): a DWI in which no gradient has been applied. At least one of these images is required as a reference for image processing.

(Note: much of this information was adapted from the Basser/Jones paper referenced below)

Note: When in doubt, 1) Check the FSL websites (listed below), 2) Check the FSL mailing list (also listed below), and 3) email me. Thanks.

1. Use dcm2nii to convert DICOM files to .nii format.
> dcm2nii dicom_directory/
dcm2nii is our preferred DICOM > NIfTI converter, because of its ability to read the gradient vectors (bvecs) and b-values (bvals) directly from the DTI DICOM files. Based on my observations, the gradient vectors from the DICOM files are accurate. In some cases, the b-value is may not be saved as 0 for the B0 images. In order to correct the b-values file, the best strategy is to view your raw DTI .nii file and determine which images are B0 images, and updates the values in the bvals file. B0 images appear brighter, the diffusion-weighted images will appear distinctly darker.
2. Eddy correct the .nii file
> eddy_correct data_raw.nii.gz data.nii.gz
3. Create binary mask
> bet data.nii.gz nodif_brain -f 0.3 -g 0 -n -m
4. Fit the tensor at each voxel using dtifit
> dtifit --data=data.nii.gz --out=dti --mask=nodif_brain_mask.nii.gz --bvecs=bvecs --bvals=bvals

Once tensors for each voxel have been calculated, a range of options are available.

Tractography
For deterministic tractography, Diffusion Toolkit and Trackvis are great programs for calculating tensors from raw DTI data (DICOM or nii), running tractography, and displaying tracks. For calculating statistics on these tracks (including connectivity matrices for network analysis), my programs in the UCLA Multimodal Connectivity Package are one good option.

For probabilistic tractography, FSL's BedpostX and ProbtrackX are a good choice.
Below is my batch for processing, from raw data to probtrackx. Refer to the probtrackx homepage.

> dcm2nii -o $workdir/$subj $dicom
> mv 1.nii.gz data_raw.nii.gz
# bvecs as read from DICOM by dcm2nii seem correct; bvals are not set as 0's for the B0 images, so use the modified bvals file instead
> rm 1.bval
> cp ../bvals .
> mv 1.bvec bvecs
> eddy_correct data_raw.nii.gz data.nii.gz 0
> bet data.nii.gz nodif_brain -f 0.3 -g 0 -n -m
> dtifit --data=data.nii.gz --out=dti --mask=nodif_brain_mask.nii.gz --bvecs=bvecs --bvals=bvals
> bedpostx $workdir/$subj -n 2 -w 1 -b 1000
> probtrackx --mode=seedmask -x mask$i.nii.gz -l -c 0.2 -S 2000 --steplength=0.5 -P 5000 --xfm=$dir/xfms/standard2diff.mat --forcedir --opd -s $dir/merged -m $dir/nodif_brain_mask --dir=$dir/network/$i --targetmasks=$dir/network/$i/targets.txt --os2t

This can be used to track streamlines between a seed mask and a set of classification masks. Once probtrackX has completed, the following commands on the seeds_to_mask* output files will count the total number of streamlines:
> fslstats $i -M
> fslstats $i -V

Group Analysis
One option is Tract Based Spatial Statistics (TBSS), FSL's package for running voxel-wise statistical analyses on various DTI-derived images (FA, MD, etc).
First, make a directory called tbss. Copy each subject's dti_FA image there. Subsequent processing will be easier if subjects in different groups are given distinguishing file names, i.e. control_subj1_dti_FA.nii.gz, patient_subj2_dti_FA.nii.gz. Make sure to refer to FSL's TBSS webpage. Below is an example of a tbss run.

> tbss_1_preproc *.nii.gz
> tbss_2_reg -T
> tbss_3_postreg -S
> tbss_4_prestats 0.2
> cd stats
> design_ttest2 design 9 12
> cd ..
> randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -n 500 --T2 -V

Basser/Jones 2002 review on diffusion MRI theory
Chris Rorden's dcm2nii page
Chris Rorden's DTI analysis page
FSL's DTI processing pipeline
FSL Archives, a wealth of information on problems faced and overcome by FSL users.