Yo all -
This is a fun conversation!
Stefano - looking at your b-values, I still think you are better to
include all the data rather than to fit to a portion of the data. The
point that is made here is exactly the same in the diffusion tensor
fit that you commonly use - a mono-exponential decay is assumed in
both models. I think Carlo has a number of reasons for preferring
this type of dataset with multiple low b-values (is it partly because
of preprocessing issues?). In any event, I suspect that you will be
sufficiently monoexponential that the bedpostx fit will be fine.
I think it is likely that Matt's problems in reconstructing second
fibres in a mulit-low-b dataset are less because of non-
monoexponential decay, and more because there simply isn't any fine
detailed information in a large proportion of the data (the lowest b
images). Hence these parts of the data, whilst useful for fitting
tensors, do not contribute to more sophisticated fits.
There is no question that if you are aiming for high sensitivity to
crossing fibres, you are better off spending more time sampling at
higher b. There is also no question that if you go to higher b and
acquire on multiple shells or on a cartesian grid the current
bedpostx model (and the diffusion tensor model and many other models)
will not be able to take advantage of all of the data. They assume a
single exponential, and the data is a long way from that (this is, of
course completely irrelevant if you only sample at one b).
The next version of bedpostx will have one significant improvement -
an initialisation procedure that dramatically improves the robustness
of second/third fibre fitting - this really makes a big difference.
The version after that will have a funky model of non-monoexponential
decay in it, so it will be applicable to any diffusion dataset that
you acquire. It is still not clear to me, however, what the most
sensitive thing will be in practice (i.e. single shell, multi shell or
q-space-style cartesian). We are going to test this (hopefully with
some help from Matt :) ).
If I were designing a new diffusion experiment to be analyzed over the
next couple of years, I would acquire on a single shell. If I had a
top-of-the range 3T a lot of time, I would probably acquire with a b
of around 2000, and I would get as many directions as I could. I
would probably acquire 2 averages with phase encode reversal and
reconstruct these images using some code that Jesper is writing, that
almost completely gets rid of susceptibility distortions. I do not
know when this code will be released to the world, but I imagine it
will be within a year or so.
If I had less time, it would depend on what I wanted to do. I would
probably go to lower b (1500?) to get some SNR back for the DTI fit,
but still give me a good chance of accurate crossing fibre
reconstructions for tractography.
Cheers all
T
On 25 Nov 2009, at 21:04, Scott Kolbe wrote:
> Hi Guys
>
> I have a couple of questions on this topic as I am contenplating my
> next set of experiments and am interested in trying to get a method
> working for tractography in patients with oedematous white matter.
>
> Matt, you said you were having trouble fitting BEDPOSTX with that b
> scheme you posted. it is my understanding that diffusion is
> adequately modelled using a monoexponential for low b-values such as
> those you used. is that incorrect and what is the advantage of using
> mutiple low b-values? would it not be better to either:
> a) collect more directional information at a similarly low b-value, or
> b) collect a wide range of b-values (0 - 10,000ish) and fit a more
> complex model or calculate a diffusion probability function?
>
> Saad, you mentioned that BEDPOST may soon be relaxed to allow
> fitting of more complicated diffusion acquisitions. when you say
> 'relaxed' do you mean that the sticks and ball might change to
> ellipsoids and ball? keep us in the loop as this sounds pretty
> exciting.
>
> cheers,
> Scott
>
>
> Marenco, Stefano (NIH/NIMH) [E] wrote:
>>
>> Average trace of bmatrix
>>
>> (sec/mm^2 )
>>
>>
>>
>> Number of directions
>>
>> 3.8
>>
>>
>>
>> 3
>>
>> 8.5
>>
>>
>>
>> 6
>>
>> 64.9
>>
>>
>>
>> 10
>>
>> 114.3
>>
>>
>>
>> 12
>>
>> 348
>>
>>
>>
>> 16
>>
>> 574.6
>>
>>
>>
>> 18
>>
>> 858
>>
>>
>>
>> 20
>>
>> 1198
>>
>>
>>
>> 22
>>
>> This is the scheme we used for thalamic segmentation, and I did not
>> notice major abnormalities here with bedpostx, however I might have
>> to look more carefully. What would be a good test of this?
>>
>> In this case, would it be advisable to drop the intermediate shells
>> (say b= 65-575?)? I thought that when we first consulted Tim about
>> this, he advised to maintain the intermediate shells to gain SNR….
>> Stefano
>>
>> ------------------------------------------------------------------------
>>
>> *From:* Matt Glasser [mailto:[log in to unmask]]
>> *Sent:* Wednesday, November 25, 2009 12:53 PM
>> *To:* [log in to unmask]
>> *Subject:* Re: [FSL] DTI with High b value
>>
>> Hi Stefano,
>>
>> What are your bvalues for this dataset? With a dataset that had the
>> following bvalues: 0 50 100 150 200 300 350 400 450 500 550 600 700
>> 700 800 850 900 950 1000 1050 1100 1150 1250 1300 1350 1400, I had
>> a great deal of difficulty reconstructing second fibers because of
>> the problem Saad describes below. Saad I haven’t tried the more
>> relaxed version on this dataset yet, but perhaps it would be
>> interesting to see what happens!
>>
>> Peace,
>>
>>
>> Matt.
>>
>> ------------------------------------------------------------------------
>>
>> *From:* FSL - FMRIB's Software Library [mailto:[log in to unmask]]
>> *On Behalf Of *Saad Jbabdi
>> *Sent:* Wednesday, November 25, 2009 11:38 AM
>> *To:* [log in to unmask]
>> *Subject:* Re: [FSL] DTI with High b value
>>
>> Hi Stefano,
>>
>> bedpostx only models one diffusion coefficient (we are testing a
>> more relaxed version).
>>
>> So if you are in a portion of the b-range where you have mono-
>> exponential decay, it's fine.
>>
>> Otherwise you will tend to have large errors in your modelling.
>> This may have some nasty effects on other parameter estimates
>> (other than diffusivity). For example, the ARD prior on the fibres
>> (which allows us to test how many fibres the signal supports per
>> voxel) may not work properly, because the extra parameters try to
>> explain the residual errors from modelling a single exponential
>> (which they can't!).
>>
>> Cheers,
>>
>> Saad.
>>
>> On 25 Nov 2009, at 15:02, Marenco, Stefano (NIH/NIMH) [E] wrote:
>>
>> Saad, I have used FSL tools with multiple b-values (within a more
>> conventional range of b<1200) and they give good results. I am
>> aware that DTIFIT will use a monoexponential decay and that the
>> extra information provided by the different b-shells will be
>> essentially ignored. Is that what you meant here? Stefano Marenco
>>
>> ------------------------------------------------------------------------
>>
>> *From:* Saad Jbabdi [mailto:[log in to unmask]]
>> *Sent:* Wednesday, November 25, 2009 6:27 AM
>> *To:* [log in to unmask] <mailto:[log in to unmask]>
>> *Subject:* Re: [FSL] DTI with High b value
>>
>> Hi Moran,
>>
>> You cannot (yet) use the FDT tools to do qspace analysis (which I
>> am not sure how you can do anyway if you only have a single b>0).
>>
>> The FDT tools for modelling the signal are DTIFIT and BEDPOSTX, and
>> those will work fine on your data (because you have a single b>0!)
>>
>> Cheers,
>>
>> Saad.
>>
>> On 25 Nov 2009, at 08:41, Moran Artzi wrote:
>>
>>
>>
>> Hi Matt,
>>
>> Thanks for your replay
>>
>> data info:
>>
>> bvalue of 12000 (non-mono-exponential decay) with 45 gradients
>> directions
>>
>> attached - bvals and bvecs files
>>
>> Many thank
>>
>> Moran
>>
>>
>>
>> On Wed, Nov 25, 2009 at 8:56 AM, Matt Glasser <[log in to unmask] <mailto:[log in to unmask]
>> >> wrote:
>>
>> FSL's model based analysis is capable of analyzing single shell
>> data of any
>> bvalue. What are the parameters of your sequence (# of directions,
>> bvalue(s))?
>>
>> Peace,
>>
>> Matt.
>>
>>
>> -----Original Message-----
>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask] <mailto:[log in to unmask]
>> >] On Behalf
>> Of Moran Artzi
>> Sent: Wednesday, November 25, 2009 12:08 AM
>> To: [log in to unmask] <mailto:[log in to unmask]>
>> Subject: [FSL] DTI with High b value
>>
>> Hi,
>> Is there a way to analyze data of DTI with High b value (q-space
>> analysis)
>> using FSL?
>> Thanks
>> Moran
>>
>> <bvasl><bvecs>
>>
>> --
>>
>> Saad Jbabdi
>>
>> University of Oxford, FMRIB Centre
>>
>> JR Hospital, Headington, OX3 9DU, UK
>>
>> +44 (0) 1865 222466 (fax 717)
>>
>> www.fmrib.ox.ac.uk/~saad <http://www.fmrib.ox.ac.uk/%7Esaad>
>>
>>
>>
>>
>>
>> --
>>
>> Saad Jbabdi
>>
>> University of Oxford, FMRIB Centre
>>
>> JR Hospital, Headington, OX3 9DU, UK
>>
>> +44 (0) 1865 222466 (fax 717)
>>
>> www.fmrib.ox.ac.uk/~saad <http://www.fmrib.ox.ac.uk/%7Esaad>
>>
>
> --
> ========================
> Scott Kolbe
> Neuroimaging Group
> Florey Neuroscience Institutes and
> Centre for Neuroscience
> University of Melbourne
> VIC, Australia, 3010.
>
> ph: +61 3 8344 1929
> email: [log in to unmask]
> website: www.neuroimaging.org.au/index.php?id=383
>
|