Hi-

I am not aware of any available software packages for the fast
marching tractography (FMT). I'm pretty sure that there is none in
AFNI, though I will point out that there is both a deterministic and a
probabilistic tractography program in AFNI. Particularly the latter
was written (disclosure: by me) with the aim of coupling DTI and fMRI
studies. It's based on an improvement to FACT tracking called FACTID,
which is discussed within the message board. And I'm happy to answer
more questions about. Briefly, 3dProbTrackID is the probabilistic
program, and it takes in one (or more) networks of ROIs, searches for
tracts which connects any pair of ROIs in a given network (and also
the set of tracts through just a single ROI), and calculates simple
stats (mean/std of FA, MD, RD and L1) on each 'WM ROI' (set of bundles
connecting a given pair of 'GM ROIs') it finds.

Related to connecting GM and associated WM regions reliably... Well,
that's a hard issue (and one which exists regardless of tractography
algorithm). Namely, from the GM location, how does one choose an
on-ramp to get onto the correct WM highway? What I've done at times
is to inflate the GM region by a couple voxels and to perform
tractography, with the aim that the inflated ROI overlaps with nearby
WM and that that WM is the physiologically correct one... I believe
that something of similar aim and sophistication (i.e., fairly simple)
is done in both the Guye et al. and Staempfli et al. papers:

Staempfli et al.: <<The seed regions for the tracking process were
defined on the basis of the COGs of activations in the left precentral
gyrus, obtained from the foot, hand, and face motor fMRI
experiments. In order to be able to start the tracking algorithm in
the adjacent white matter, the COGs were enlarged spherically.
Thereby, all adjacent voxels whose midpoints were less than 3 mm away
from the COG were included, resulting in a volume of 19 voxels
(Staempfli et al., 2004). Due to the low anisotropy inside the
striatum andwithin the starting regions in the precentral gyrus (Table
1), the tracking algorithm’s fractional anisotropy stop criterion was
decreased from 0.2 to a value of 0.1 in these areas.>>,
where I believe COG='centre of gravity'.

Guye et al.: <<The coregistration of fMRI and DTI images allowed us to
select the seed points on the FA maps within the white matter directly
adjacent to the activated precentral gray matter during fMRI. We
selected a set of 3 adjacent seed points in each hemisphere, focused
on the highest significantly activated fMRI voxel (Fig. 1). The
selection of three seed voxels was based on the observation that
connectivity maps resulting from neighbouring voxels often result in
markedly different patterns of connectivity subsampling different set
of cortical connections (as is to be expected in light of the complex
nature of cerebral interregional connections) (Ciccarelli et al.,
2003). The selection of three adjacent seed points therefore reduced
the possibility of missing important pathways of connection due to
small errors in seed point placement (Fig. 2).>>,
where I believe that WM='matter with FA>0.1' (a la their comment, <<We
excluded from the FMT calculation, the voxels with a FA value lower
than 0.1 which excluded artifactual paths through CSF but allowed the
algorithm to make cortico-cortical connections.>>)

Now, I note that FA~0.1, which both papers and probably FMT algorithms
in general use, is quite low, and if you look at pictures of an
ellipsoid of that low FA, it might be surprising how nearly spherical
its shape is. I think that ellipsoids with low FA values (even up to
0.2) are quite susceptible to noise effects, such as rotations away
from alignment with physical tract fiber bundles, which I think that
can be shown easily with simulations or uncerainty measures of real
data; therefore, I would wonder a bit about using low-FA eigenvalues
to 'point the way' to real WM. At least for fast marching, having an
area of GM/low-FA voxels should mean that the propagation front is
pretty uniform, i.e. spherical, so that I would think that enlarging
an ROI or making a sphere of reasonable size around it would be a fine
method for finding nearby WM; it's a simple and somewhat reasonable
way of going, and I don't know that fancier methods in practice can
prove themselves better. One way of verifying a 'good' choice of WM
tract near an ROI might be to see if it appears to be 'pointing'
toward the region within the voxel (i.e., e_1 orientation points
mostly toward ROI), but that is quite problematic in practice due to:
noise, uncertainty cone of a given e_1 can be large by definition for
low-FA values, unknown physiology of whether a tract *really*
approaches so directly, etc.

--pt