DTI Scalar feature

[synic]

Dear experts,

I'm very interested with the scalar feature provided for the tracks.
1) I would like to know if there is a published reference we can cite if we want to use it.
2) I don't understand what exactly the scalar feature is doing in the case of the measurement of a "Y" shaped track

thank you for your time,

cheers

michel

[Yann Cointepas]

Some features just consider a bundle as a set of 3D points. These points are projected into several images (FA, ADC, VR, etc.) and for each image a few statistical values are extracted (max, min, mean, standard deviation, etc.). For Y shape bundle, I suppose you are interessed in values extracted along a bundle. The algorithm is too simple to work on bundles without a "good" geometry. A good geometry means all starting points are close together, all end points are close together and the bundle contains almost parallel curves. Here is a summary of the algorithm:

1) Identification of curves orientations in the bundle
starting_points = empty set of points
end_points = empty set of points
starting_point_reference
end_point_reference
for each curve in bundle:
if starting_points is empty:
starting_point_reference = first curve point
end_point_reference = last curve point
add first curve point to starting_points
add last curve point to end_points
else:
if first curve point is closer to starting_point_reference than to end_point_reference:
first_point = first curve point
last_point = last curve point
else:
first_point = last curve point
last_point = first curve point
endif
add first_point to starting_points
add last_points to end_points
starting_point_reference = mean( starting points )
end_point_reference = mean( end points )
endif
end for

2) Cut the bundle in N "slices"
For each curve, N points are equaly distributed on the curve. The first one is the first point of the curve, the Nth one is the last point of the curve.
This defines N set of points. The set number I contains all points of index I for all curves. For each set, the same statistical values as for the whole bundle are computed. These results is displayed as curves in BrainVISA.

There is no publication because I am not sure this algorithm is not robust to bundle gemometry. I am not sure it deserves a publication. People in our lab have developped good methods to extract the central line of a bundle. Unfortunately we have not yet finished the job of plugging this in BrainVISA. There are also a few command lines in BrainVISA package that do not have a BrainVISA process to use them through a graphical interface. For instance, with comistBundle2Ima, you can create an image from bundles (curves density map). It is interesting to use classical image processing tools to analyse tracking results.