[Mrtrix-discussion] streamtrack SD_PROB

[Donald Tournier]

Hi Luke,

In answer to your questions:

> I'd like to get some more information about the method that you use for your
> probabilistic tracking using the fods, specifically what it means to use an
> inclusion/exclusion roi.

An inclusion ROI means that only tracks that go through it will be
written to file - any tracks that don't go through the inclusion ROI
will be discarded. An exclusion ROI is means that any tracks that go
through it will be discarded immediately.


> 1) the progress bar for streamtrack is generally something like this 'N
> generated,      X selected', what does this mean? what is the selection
> criteria, I get this message even if no inclusion or exclusion rois are set.

By default, streamtrack will keep generating streamlines from the seed
ROI until the requested number of streamlines have been selected (i.e
written to file). This means that it will generally need to generate a
much larger number of tracks than are actually selected, since a large
proportion of the tracks will be discarded. You will get this message
every time you run streamtrack, whether you have addition
inclusion/exclusion ROI or not - it's just the default behaviour.

Note also that there is a minimum length parameter (by default, this
is 10 mm) - any tracks shorter than this will be discarded. So even
with no inclusion/exclusion ROI, the number of tracks generated will
generally be greater than the number selected, simply because a
proportion of the tracks generated will be too short.


> 2) Lets say that I set an inclusion roi, is the process to generate fibers,
> until enough get to the inclusion roi? if so it seems as given enough time
> I'd be able to fiber track anywhere to anywhere. Am i missing something?

You're not, this is exactly what happen. There is however a limit on
the maximum number of streamlines generated. By default, this is 100
times the number requested. This limit can be set manually with the
-maxnum option if that is more appropriate for your application. In
particular, if you set both -maxnum and -num to the same number, you
can get streamtrack to generate that number of streamlines, regardless
of how many actually get selected.


> 3) Is there anyway to generate confidence intervals on the fibers that are
> generated? If I'm using an exclusion or inclusion roi, I'd like to have an
> idea with how likely a fiber is to.

If you're using the probabilistic streamlines algorithm (which I'd
recommend you do), then the results themselves provide a kind of
measure of confidence. The typical way of interpreting the
probabilistic results is that the density of the tracks in some way
reflects the likelihood of a connection.

You might have been asking a different question though. If you want to
have the 'probability' of a connection, then what you need is the
fraction of the total number of streamlines generated that were
actually selected. You can get this either by setting the -maxnum and
-num options to the same value (as above), and/or by reading off these
numbers from the track file header (using the track_info command).

Hope that answers your questions.
Best regards,

Donald.


-- 
Jacques-Donald Tournier (PhD)
Brain Research Institute, Melbourne, Australia
Tel: +61 (0)3 9496 4078