A set of diffusion tensor (DT)-based deterministic and probabilistic
tractography programs are now available for use in AFNI. These include:
- deterministic tractography (3dTrackID),
- probabilistic tractography (3dProbTrackID),
- and a program to estimate DT parameter uncertainty, for use with probabilistic tractography (3dDWUncert).
These have been variously developed, code/implemented and tested by
Paul Taylor, Kuan-Hung Cho, Ching-Po Lin, Bharat Biswal and Ziad Saad.
These were added on the weekend of July 1, 2012, and can be found in
the updated AFNI binaries distribution. If you have an AFNI version
postdating prehistoric times, simply run the following for an update:
@update.afni.binaries -d
These programs should interface easily with AFNI processing or with
other DT-estimating softwares, such as with the TORTOISE pipeline (via
the NIH Pediatric Neuroimaging Diffusion Tensor MRI Center; see e.g.,
Pierpaoli et al., 2010). Visualization can be performed using AFNI,
SUMA, and the freely available TrackVis (Wang et al., 2007).
Also included in the update is a simple set of demonstration files and
scripts for the varied programs. To get and run the demos, execute the
following command after the update command above:
@Install_PTaylor_TractDemo
This script will download and unzip the directory `PTaylor_TractDemo'
in whichever directory you run this command. In the created
PTaylor_TractDemo directory is a set of diffusion weighted data, ROI
masks, a README.txt text file, some supplementary files, and a series
of executable scripts (Do_0*, simply enter their names in the
commandline) which first run relevant AFNI programs to calculate DT
parameters, and then sequentially run and view deterministic
tractography and probabilistic tractography.
We note that the tractography itself is powered by streamline
tracking, using an improvement to FACT (Mori et al. 1999) which we
term `FACTID' (FACT including diagonals). Compared to FACT,
propagation of tracts via FACTID is less biased by the coordinate axes
of scanning and less sensitive to noise in data, as shown in a poster
at the recent OHBM conference (#424) and in a more complete study by
Taylor, Cho, Lin and Biswal (2012, submitted) (and FACTID compared
favorably with other software and methods of tractography, as well,
such as DTI-Studio FACT and DTI-Query 4th order Runge-Kutta
integration).
In 3dTrackID, one may use tractography to estimate tracts through
either one or two (using AND/OR logic) regions of interest (ROIs),
with each region simply being a binary mask data set. Outputs include:
- TrackVis-readable file for viewing the calculated tracts themselves,
- An AFNI `map' file containing the count of how many tracts passed through a given voxel,
- An AFNI `mask' file containing a `1' in voxels through which at least one tract passed and a `0' in others,
- A SUMA-readable file to view the tract results in a 3D brain.
3dProbTrackID was written with a view toward complementing studies of
networks within the brain: finding white matter (WM) regions of
interest (ROIs) associated with network connections and calculating
statistics of them. Therefore, one may input a network of interest,
e.g., the default mode network (DMN) from those popular resting state
studies, to test the WM-based connections between the four distinct
gray matter (GM) regions: one would great a mask defining each sub-ROI
of the network with an integer: 1, 2,..., 4 in this case.
3dProbTrackID then performs Monte Carlo simulations of whole brain
tractography using uncertainty estimates of each diffusion tensor's
first eigenvector and FA (from jackknife resampling with 3dDWUncert),
keeping track of the connections through each individual ROI and
through each possible pair of network ROIs. After thresholding, one
obtains a map of WM ROIs associated with each individual ROI and each
possible pair. Statistics (mean and standard deviation) of DT
parameters (FA, MD, first eigenvalue L1 and radial diffusivity
RD=[L2+L3]/2) are automatically calculated and returned in a matrix
(in this case, 4x4 since there were four GM ROIs). It is also
possible to test connections between several networks simultaneously,
with each network input as a sub-brick of ROIs numbered with integers
1,2,...,[total number of network ROIs].
The output *_NETMAPS.BRIK file of 3dProbTrackID formatted as a set of
subbricks as follows (we note that to label connections between ROIs
which might contain overlapping regions, we had to use binary indexing
as a unique means of decomposing results):
- subbrick [0] shows info on tracts related to the whole network: a binary mask of all voxels contained in tracts through any network ROI (OR logic) after thresholding.
- subbrick [1] shows info on tracts related to ROI-1:
- voxels in tracts that passed through ROI-1 only have value 2=(2^1),
- voxels in tracts that passed through both ROI-1 and ROI-2 only have value 6=(2^1+2^2),
- voxels in tracts that passed through both ROI-1 and ROI-3 only have value 10=(2^1+2^3),
- voxels in tracts that passed through both ROI-1 and ROI-2 and both ROI-1 and ROI-3 have value 14=(2^1+2^2+2^3),
- ...
- subbrick [2] shows info on tracts related to ROI-2:
- voxels in tracts that passed through ROI-2 only have value 4=(2^2),
- voxels in tracts that passed through both ROI-2 and ROI-1 only have value 6=(2^2+2^1),
- voxels in tracts that passed through both ROI-2 and ROI-3 only have value 12=(2^2+2^3),
- voxels in tracts that passed through both ROI-2 and ROI-1 and both ROI-2 and ROI-3 have value 14=(2^2+2^1+2^3),
- ...
- {and so on, up to the Nth sub-brick representing the total Nth sub-ROI of the input network}.
Additionally, a *.grid file is output per network, containing
grids/matrices of statistics of the found WM-ROIs, with number of
tracks found and mean/std of FA, MD, L1 and RD. We view the main
purpose of probabilistic tractography to identify regions of WM likely
to be associated with GM or other ROIs, and to quantify properties of
those WM regions for complementary use to, for example, GM-based
quantities such as gray matter volume, correlation coefficients,
network properties, etc. We do not recommend treating `number of
tractographically returned tracts' as a `strength of structurally
connectivity' parameter (number of tracks is returned just as a number
of minor information), as it depends quite strongly on numerical
features involved and is likely unrelated to the physiological
quantity of number of fibers or `strength of physical connection'
(whatever that might mean differently to different people, anyways).
These steps are all demonstrated in the available demo. These
programs have been used on several data sets, but additional feedback
would be useful, either in the form of questions, complaints or
compliments. Furthermore, comments assist our aim for these programs
to develop further both in terms of expanding applications and in
algorithm improvement. Questions could be posted to the AFNI
community page or emailed to Paul Taylor (neon.taylor _at_ gmail.com).
Edited 1 time(s). Last edit at 07/06/2012 10:32AM by ptaylor.
