2) Aug 30, 2012
Executive Summary:
We discussed the choice of scanner parameters for DTI, tractography and TBSS.
More Detail:
Scanner Protocols: Dianne reviewed the protocol she used for DTI acquisition and analysis (posted below).
Stephen Wilson and Ted Trouard provided insight and detail about important fundamental concepts.
Anisotropy: Ted explained more about why anisotropic voxels are a bad idea in dti...not just because of
issues with vector reconstruction, but also because anisotropic voxels would cause orientation-specific
partial voluming of scalar measures like Fractional Anisotropy.
B-values and Gradients: Ted explained B-weighting, and promised to provide greater detail next time.
There is considerable variability in the extraction of bvecs files (a text file of the gradient orientations for each direction).
Some programs dcm2nii, seem to extract exactly the bvecs that FSL wants. Other programs, like MRIconvert, extract
the bvecs as they are originally represented on our scanner. In the latter case, the sign of the Z gradient is
reversed relative to what FSL expects. It is exceedingly important to look at the primary vectors generated by your
combination of scanner, reconstruction program and display/analysis program.
Registration: We discussed registration of DTI images (they are distorted because they are EPI images)
eddy_correct in FSL reduces some of the distortions and does realignment of the DTI volumes:
"Eddy currents in the gradient coils induce (approximate) stretches and shears in the diffusion weighted
images. These distortions are different for different gradient directions. Eddy Current Correction
corrects for these distortions, and for simple head motion, using affine registration to a reference volume."
http://www.fmrib.ox.ac.uk/fsl/fdt/fdt_eddy.html
We discussed image registration more generally, and Stephen mentioned being impressed with 2 tools for registration.
ANTS: http://www.picsl.upenn.edu/ANTS
and
DARTEL (part of SPM): J. Ashburner. A Fast Diffeomorphic Image Registration Algorithm. NeuroImage, 38(1):95-113, 2007.
(Ashburner_2007.pdf attached below)
Skull Stripping: For manual mask editing, Dianne mentioned IMANGO, an ipad app ($14.99):
http://ric.uthscsa.edu/mango/imango.html
What's nice about imango is the ability to work in comfort with a very easy interface.
When automatic skull stripping algorithms fail, this is a reasonbly pleasant alternative.
(The free MANGO http://ric.uthscsa.edu/mango/mango.html is also available for mac, windows or linux,
but, of course, lacks some of the ease of working on the ipad touch screen.)
Tractography: Dianne and Stephen both do tractography in native space. Interestingly, the FSL group gently
recommends tracking in standard space:
A better way to perform tractography [is] directly in standard space, providing probtrackx with the
transformations between diffusion and standard space. In that case, the binning of the spatial histogram
will be done directly in standard space, and you don't need to worry about the interaction between local
contraction/expansion and interpolation.
The strength of probabalistic tractography in FSL is that it can follow branches of a tract, and provide an
estimate of the likelihood that a tract follows a particular path. Just like with fmri, you must decide on the appropriate level
of thresholding, and apply your scheme consistently.
Here's a sample of an arcuate tracking: An exclusion mask prevents the algorithm from crossing the corpus
callosum or passing through the insula. On the left is the unthresholded result, on the right,
some thresholding has been applied:
FSL also generates and uses information about crossing fibers if there are enough directions in the dataset to
do so. This crossing fiber information is used in the tractography. In fact, probabalistic tractography can penetrate
pretty deeply into grey matter at the ends of tracts. This turns out to be very useful for parcellation using classification targets.
This ability to penetrate into regions of reduced FA is a weakness too: probabablistic tractography can get out
of hand and track through regions of CSF or create phantom connections.
TBSS: Stephen followed up by talking informally about TBSS (tract based spatial statistics):
http://www.fmrib.ox.ac.uk/fsl/tbss/index.html
TBSS is like VBM (Voxel Based Morphometry) for white matter and is part of the FSL package.
It is designed for doing whole brain comparisons of subject groups to identify differences in the major tracts.
Ted observed that because TBSS works with an eroded skeleton, it might only be sensitive to changes in the core of the tract.
Stephen Smith of the FSL group confirms that this is true, and that, at least for the thicker tracts, this means that TBSS would not
detect changes on the periphery of the tract. However, he notes that for tracts that are sufficiently thin, you might see a change
even on the skeleton "which is a good thing (sensitivity) and a bad thing (interpretability)" (Stephen Smith, personal
communication, 9/1/2012).
If Oeschlin 2010 (attached below) is correct about the "pioneer axon thesis", then TBSS may be especially sensitive to
changes in those axons which were laid down earliest in development (in the core of the tract), rather than axons laid
down later on the periphery. Oeschlin hypothesizes that peripheral axons are most influenced by environmental factors.
Other Packages of Interest: Kyle Almryde mentioned that Afni now has a probabalistic DTI package:
http://afni.nimh.nih.gov/afni/community/board/read.php?1,86370,86370#msg-86370
Kyle also mentioned that AFNI also has rat and monkey brain templates.