Ok .... I'm a MD and I'm sure I'll be less clear and exhaustive of any biomedical engineer in this group ... but let me try!!
Because voxels are much larger than
axons, each point in the brain does not have just one connection, but
many. Furthermore, because diffusion measurements contain errors the reconstruction of a tract can never be perfectly accurate, even if there were only one connection we would always be uncertain as to its exact pathway.
More simply .... Deterministic Tractography estimates at every voxel the “most likely” fiber orientations and this means it has an intrisic grade of uncertainty. Probabilistic Tractography has been developed for trying to overcome this problem and limiting the uncertainty. It estimates a distribution representing "how likely" each other
orientation is to lie along a fiber. It traces the
connections thousands of times, each time using slightly
different orientations (according to their likelihood). The set of all
these different paths are then collectively a measure of the connection
likelihood or probability.
In this sense the Deterministic Tractography is alot more complicated. It isn't a case my system needs few minutes for the Deterministic and 4-5 hours for Probabilistic Tractography. Talking about this I found very fantastic the new "parallel computing" recently introduced in FSL. It reduces the time of the computing hundreds of time.
For more clear explanation of advatanges / disadvatanges of these two methodologies my advice is to read this
paper (in particular pages 10-11).
While Deterministic Tractography is possible with a large number of tools (TrackVis-FSL-Camino-DTI Studio .... etc.) I know only 2 tools where the more complicated Probabilistic Tractography is possible: 1)
FDT (part of FSL) ; 2)
Camino .
I'm sure reading the wiki of FSL and Camino you can find alot of interesting infos.
Ok, I hope this can help you. I'm very curious to know the opinion of others on this matter.
Best regards,
Alessandro