Dear Tim,

Regarding to your reply to James' questions:
1.what is the efficient way for examining FA along the pathway to detect
whether it's the cause of uncertainty?
2.could you give me some suggestions to follow up the reasons of change
in probability?

Many thanks for your help,

Shiva

-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
Behalf Of Tim Behrens
Sent: 09 July 2008 09:23
To: [log in to unmask]
Subject: Re: [FSL] Volume, probability, and intensity questions.

Dear James -

Sorry for the delay.

In general our advice is this -

1) Comparing probabilities should be done with great care - there are
really only a two or three of reasons why probabilities might be
different.
i) - a diffuse reduction in FA along a whole pathway will add
uncertainty and lower probabilities - This should be detectable by
examining FA along the pathway, and will be much more interpretable in
this fashion.
ii) A focal change in FA may cause a focal spread, which is likely to be
detectable with probability, but not group FA comparisons (as this focal
change is unlikely to align across subjects).
iii) A focal change in the orientation of diffusion across groups - this
might be caused by changes in routes taken by pathways between the
groups.

If you see a change in probabilities, it is therefore essential to
perform many follow-up studies to confirm the exact mechanism causing a
change in probability.


2) Using tractography to generate ROIS.
We have not had conceptual problems with this idea - It should be
possible to generate consistent ROIs using probtrackx outputs, with
which you can later measure the volume, mean FA etc. etc.

In order to do this you should be able to use a mixture of waypoint
masks, and termination masks to generate consistent ROIs. We have found
that this works very well.
In order to compute the volume, we would usually play with these
different masks as much to isolate the relevant pathway as well as we
can. We would then threshold at a low level to remove the noise, and
binarise.
We have found that the thresholding technique that is most consistent
across subjects is to threshold using a percentage of the total number
of pathways that made it from seedmask to waypoints. This number is
stored in waytotal - however, if you have been following recent emails,
you will know that the current version of probtrackx has a slight bug in
the computation of waytotal. However, the number you get out will still
be approximately proportional to the right number, so it should still be
fine to threshold at a consistent percentage of this number.

I will try to answer some of your more specific Qs below.


> My thesis advisor and I are wondering if it is even valid to compare
> mean FA, MD, and Volume when using: 2 roi's, an exclusion mask, and
> probability tracking. If one subject has a many fibers connecting the
> two roi's while another subject has many that do not connect and only
> a few that do, then the probabilities will be weighted very
> differently.


You are right that you have to be able to track the pathway robustly in
each subject.
If you cannot, then it is a bit of a non-starter. If you can, however,
then, if you threshold as described above, there is no reason I can see
that will make you biased towards subjects where the pathway is easier
to track, as this will also be visible in the number in waytotal .


> Are these probabilities
> subjective and not comparable. Does this make comparisons between
> groups when using 2 roi's not valid?

I think they are comparable - see above.


> How does an exclusion mask effect probability when a tract hits the
> exclusion is it removed from the probability calculation?
>


An exclusion mask will not count any streamline that passes through it.
A termination mask will count it and then stop it.

> I am still wondering about the questions below(from july 7th) as well.
>
> Perhaps, my question should be... what is the best way to evaluate
> group differences in volume, FA (TBSS I'm guessing), and MD on a
> specific tract?
> What is probabilistic tractography best used for? Can I use it for
> group comparisons when trying to connect 2 regions of interest?
>


  TBSS and tractography-based analysis are complementary. We find that
it is very useful to do both - they are sensitive to different things.

see above for discussion of 2 mask tractography analysis - we, and many
others, find that this is a very sensitive way to define ROIs for
further analysis.


> Any enlightenment on this would be awesome.
>
> Thanks,
>
> _J Sheehan
>
> On Mon, 7 Jul 2008 03:39:45 +0100, James Sheehan <[log in to unmask]>

> wrote:
>
>> Hey all,
>>
>> just a few quick questions.
>>
>> 1. After running a few probtrackX with 2 roi masks on different
>> subjects. I am wondering what the intensities mean. I have a few
>> tracts which are 1000+ max and some which are as low as 1 max. With 2

>> masks are these numbers more subjective? What concerns should I have
>> about the tracts with max intensities of 1?

As above, if you threshold based on a % of the waytotal, this effect
will be removed except in extreme cases.
You do have to be able to track the pathway robustly though, so if you
have a max of 1, you are always in trouble.
I would think you want a max of greater than 100 before you can
realistically be confident.


>>
>> 2. Is the Volume calculation dependent on the probability of voxel to

>> the
> tract?

No, if you follow the instructions above.

>>
>> 3. How is probability calculated with two masks, is it dependent
>> on the roi
>> and seed points, or on the number of total tracts found to the
>> number that
>> reach the 2nd roi, or something completely different?
>>


Total number of pathways that pass through a voxel and pass through
waypoints, and do not pass through exclusion ...

>> Thanks so much for any light that you can shed!!
>>

Hope this helps - sorry (everyone) that I am slow at the moment -
lots on.


Cheers

T



>> -J
>> =====================================================================

>> ===
>