I've certainly found both Mark and Tim to be extremely helpful over the
years and have learned an enormous amount from both of them. I know it is
greatly appreciated when support requests are directed to the public
e-mail list (so that others may benefit from and participate in the
discussion).
The relationship between the number of streamlines sent out from an ROI
that reach another and the number (let alone mention functional influence)
of axons is not yet well characterized. It's worth keeping in mind that
FA (and more advanced measures like f1, f2, and f3) do not have any direct
influence on tractography. Tractography is based on sending out some
number streamlines and following samples of fiber orientation
distributions until some stopping criterion is reached (like a stop mask,
exceeding a curvature threshold, looping back on a region previously
traveled, etc). Streamlines do not increase in number depending on the
number of steps (i.e. distance) the tractography algorithm needs to take
between seed and target. One starts with a fixed number (that you set
when running probtrackx) and this represents the maximum possible value
you could get (if all streamlines travelled the whole distance between
seed and target ROIs).
The two main applications where it has been successfully used are to build
spatial uncertainty distributions of pathways (where the voxels with the
largest number of streamlines represent the higher probability of a
pathway being located there and the voxels with fewer numbers of
streamlines represent lower probability of a pathway being located there.
The other is making comparisons between streamline counts in the same
individual, where many uninteresting factors that influence streamline
counts are controlled for.
The use of tractography to produce grey matter to grey matter structural
connectomes is a relatively recent one. Interpreting these results across
individuals or between patients and controls relative to the various
potential confounds is quite challenging. In this case, the explanation
could be as simple as the subjects with neurodegenerative disease have
smaller (atrophied) brains and therefore the distance between ROIs is
reduced, which would increase streamline counts because streamline counts
decrease with increasing distance in a log-linear fashion (more steps
means more chances to be stopped/diverted away from the target).
Peace,
Matt.
On 1/18/14 7:30 AM, "Meoded, Avner (NIH/NINDS) [E]" <[log in to unmask]>
wrote:
>Dear Mark,
>
>I see you like to make assumptions. So let us assume that I am an high
>school student who does not understand a thing about DTI. Moreover I do
>not
>understand the basics of probabilistic tractography. So in order to
>understand more I read Diffusion MRI book (edited by Berg and Behrens) and
>also papers that deal with different tractography methods. So now I
>understand a little bit of tractography but still there are unclear issues
>that I would like to clarify with the experts in the field- that is the
>reason I contacted FSL community.
>If you see the title of my mail it is "Values in matrix1 after
>probtrackx";
>The specific question I have is what those values mean? Probabilistic
>tractography aim to quantify uncertainty on the PDD and build a
>connectivity
>distribution. Now if you go and check matrix obtained from network1 option
>in probrackx2 you will see that the matrix contain values in the range of
>1-1,000,000 and beyond. You mentioned in you last e-mail that:
>"Uncertainty
>in direction at any point in the brain will enhance the uncertainty in the
>tractography from that point onwards for any tracks that pass through that
>point." How can we learn about this uncertainty from the matrix values?
>
>Indeed at the NIH there are many experts who are always available for
>discussion about all aspects of health and science. However, is
>FSL/Probtrakcx a NIH application???
>
>Finally I would like to show Timothy Behrens's response to my question :
>
> "if you have posted it to the FSL list then you should get an answer
>soon.
>It is a very effective community forum. You will understand that with
>more
>than 5000 users, there is no way I can personally answer every question
>and
>hope to maintain a research career!"
>
>This is the answer from the researcher who is the first author on the
>NeuroImage paper from 2007 about probabilistic tractography, and also the
>one who wrote the chapter MR diffusion tractography with Saad Jbadi in the
>book mentioned above.
>
>
>
>
>Avner
>
>
>On 1/18/14 4:58 AM, "Mark Jenkinson" <[log in to unmask]> wrote:
>
>> Hi,
>>
>> Artifact "correction" methods don't fully remove all artifacts, so you
>>cannot
>> rule out the possibility that artifacts are causing the things you are
>>seeing
>> just because you have run artifact correction. Such methods remove a
>>lot of
>> the effect of artifacts but not absolutely everything.
>>
>> I'm not sure what you mean by "steps" but the samples in probtrack
>>refer to
>> individual streamlines (that are selected from the probability
>>distribution of
>> possible streamlines). Uncertainty in direction at any point in the
>>brain
>> will enhance the uncertainty in the tractography from that point
>>onwards for
>> any tracks that pass through that point.
>>
>> You definite cannot make categorical statements such as "more samples
>>mean[s]
>> more disease".
>> As I said, there are a *lot* of things that can influence tractography
>>results
>> and you really should discuss you particular case, with your particular
>> subjects and you particular data acquisition, with someone who is
>>experienced
>> with tractography. There certainly should be such people in the NIH.
>>
>> All the best,
>> Mark
>>
>>
>> On 17 Jan 2014, at 13:10, "Meoded, Avner (NIH/NINDS) [E]"
>> <[log in to unmask]> wrote:
>>
>>> Hi
>>> The raw DTI data was corrected for artifacts.
>>> As you mentioned less uncertainty may enhance measures of
>>>connectivity. But
>>> in my case I documented reduced FA and not increased FA, the latter is
>>>seen
>>> perhaps in regions with reduced crossing fibers.
>>> Now my question is specific to probabilistic tractography: number of
>>>samples
>>> obtained from probtrackx between to regions mean number of "steps"
>>>track
>>> does; are those "steps" depends on the uncertainty? So at the end more
>>> samples mean more disease?
>>>
>>> Thank you
>>>
>>> Avner
>>>
>>> On 1/17/14 7:39 AM, "Mark Jenkinson" <[log in to unmask]>
>>>wrote:
>>>
>>>> Hi,
>>>>
>>>> These are not simple questions and it will depend a lot on the nature
>>>>of
>>>> your
>>>> data - SNR, artefacts, amount of movement, etc. There are also some
>>>> potential
>>>> biological possibilities, such as reduction in crossing tracts, which
>>>>can
>>>> enhance measures of connectivity (since there is less uncertainty in
>>>>the
>>>> crossing region) without meaning that the axonal tract is biologically
>>>> "stronger". You should look very critically at your data and show it
>>>>to
>>>> people who are experienced with diffusion analysis.
>>>>
>>>> All the best,
>>>> Mark
>>>>
>>>>
>>>> On 16 Jan 2014, at 18:08, "Meoded, Avner (NIH/NINDS) [E]"
>>>> <[log in to unmask]> wrote:
>>>>
>>>>> Dear FSL users
>>>>>
>>>>> I conducted a study with network1 option and then did structural
>>>>>connectome
>>>>> analysis, in patients affected with neurodegenerative disease.
>>>>> I also performed TBSS and found reduced FA values in different areas
>>>>>in
>>>>> patients compared to controls.
>>>>> The problem is that I have higher values stored in the connectivity
>>>>> matrices
>>>>> in patients compared to controls, and hence after connectome
>>>>>analyses I
>>>>> obtained networks that are more connected in patients. Now I know
>>>>>that
>>>>> these
>>>>> values cannot represent axons, but how you can explain reduced FA in
>>>>> patients
>>>>> and more streamlines evaluated in probtrackx? Or what are the
>>>>>numbers
>>>>> stored
>>>>> in the matrix mean? (after running seed to seed network 1)
>>>>>
>>>>> Is this because in patients (with white matter disease, lower FA)
>>>>>there is
>>>>> more uncertainty in voxels between roi1 and roi2 and therefore we
>>>>>get more
>>>>> samples so basically tracts tend to spread more and as a results more
>>>>> sample?
>>>>> so at the end more samples which represents more uncertainty
>>>>>(disease)
>>>>> Should I normalize the matrices in some way
>>>>>
>>>>>
>>>>> Thank you
>>>>>
>>>>> Avner
|