Diffusion tensor imaging (DTI) provides information on the self-diffusion of water. In regions without boundaries, diffusion occurs equally in all directions (isotropy). However, when boundaries to this free diffusion are present (e.g., due to cell membranes, myelin sheaths), diffusion tends to be greater along the long axis of those boundaries (anisotropic). In general, anisotropic diffusion has been modeled as an ellipsoid.
Diffusion tensor imaging is typically implemented using echo planar imaging. At minimum, at least seven images must be acquired at each image slice: an image with no diffusion weighting, and images with diffusion weighting in 6 different noncollinear directions. From these images, the 3 principal eigenvalues and eigenvectors of the diffusion ellipsoid can be derived. Diffusion anisotropy measures can be computed from these eigenvalues and eigenvectors.
The newest sequence in use at Nathan Kline Institute acquires 8 diffusion-weighted images, along with a non-diffusion weighted image, of each of 19 slices. There are 7 repetitions of each slice, for a total of 19x9x7 = 1197 images. This scan is acquired in the axial plane parallel to the bicommissural plane, takes 6 min, 18 seconds, and yields a signal to noise ratio (SNR) of at least 60.
The most commonly used measure of diffusion anisotropy is fractional anisotropy (FA). This measure can vary from 0 to 1, with higher values reflective of greater diffusion anisotropy. Because of the role of myelin in facilitating neurotransmission, and its role in facilitating diffusion anisotropy, studies of DTI have tended to focus on understanding diffusion anisotropy in white matter. Reductions in FA have been found in neurological conditions involving white matter pathology (e.g., multiple sclerosis). In addition, they have been found in cases of head injury, and in psychiatric disorders including schizophrenia, geriatric depression, and cocaine dependence. Thus, DTI appears to be a useful tool in examining white matter integrity.
As noted above, however, DTI generates a number of images. Babak Ardekani, PhD, has developed a software platform to generate usable data from these images. The following describes these steps.
After scanning, data are downloaded from the Siemens console to the Linux system. Ideally, DTI data should be placed in a separate directory. The computation of FA involves two steps.
This can be accomplished by using the vision2raw program:
vision2raw -c -o <output file prefix> <input Vision files>
The -c is optional, but is used to concatenate all of the Siemens files into one large raw file. This command generates a file with a .raw extension. This command is suitable for data generated using the older DTI sequences available on the Siemens scanner (in which data are averaged at the scanner). However, for newer sequences, this averaging is not done. Thus, an alternative command can be issued for these newer sequences:
avgvision -nt <number of time frames> [-tz] -prefix <output prefix> <input Vision files>
For this command, -nt refers to the number of repetitions, and -tz is used if the time frame is faster than the slice index (true for our 8 direction sequence).
After this command is issued, a *.raw image will be generated for each slice x repetition. These can be combined using:
cat *.raw > <prefix>.raw
where <prefix> can be of your choosing.
A final consideration for this step is that image names may have varying numbers of digits. Thus, if the first image is, e.g., 171-1-1.ima, the following programs would read the next image as 171-1-10.ima, and then 171-1-100.ima. This kind of reading will result in erroneous data. To address this problem, the command:
renamevision *.ima
can be issued prior to the avgvision command.
To compute FA and other relevant measures, issue this command (all on one line):
dti -nx <nx> -ny <ny> -nz <nz> -o <output file prefix> -i <input file prefix> [-gvfile <gradient vectors file>] [-T <percent of maximum>] [-TIFF] [-lambda] [-mask <mask file>] [-tz] [-P1] [-dx <x voxel dimension>] [-dy <y voxel dimension> ][-dz <z voxel dimension>]
where
-gvfile refers to the x, y, z vectors of diffusion gradients, -TIFF results in the generation of "color maps" of the eigenvectors with RED being left-right, GREEN being anterior-posterior, and BLUE being superior-inferior directions, -lambda provides images for each eigenvector (in floating point format), -trace provides images of the sum of eigenvalues for each voxel, -P1 provides images of the x, y, and z components of the principal eigenvectors -T is used for thresholding the image in order to find a brain mask -mask allows the user to provide his/her own brain mask -dx, -dy, and -dz allow the user to specify voxel dimensions -tz is used if the time index varies faster than the slice index. (Note: this option is not needed if the -tz option is specified in the avgvision step described above).
For more information, try the dti manpages:
man dti