[Mrtrix-discussion] dwi2response fsl error

[Robert Smith]

Hi Romain (and all),

Thanks for giving me access to you data so that I could investigate the
problem.

So as Donald said previously, there's been a number of issues arise
regarding the dwi2response command in the new MRtrix. This program is
almost identical to the method described in this paper
<http://www.sciencedirect.com/science/article/pii/S1053811913008367> (with
a few tweaks), which is recommended reading for anyone interested in
spherical deconvolution. We are looking into alternative ways to estimate
the response function, but these are not likely to appear overnight.

I have tweaked a couple of the default parameter values in dwi2response,
which should prevent the program from running for long periods of time. I
had disabled certain mechanisms for terminating the optimisation, having
found them unnecessary for my own data, but clearly they are necessary in
other circumstances.

However, it is important to consider the relationship between the data you
pass to this process, and the parameters you select. Of particular note is
the maximum harmonic degree used in the response function estimation. By
convention, if you do not specify *lmax* explicitly, MRtrix will select the
largest *lmax* that is supported by your image data, based on the number of
unique diffusion sensitisation directions in the b-value shell being used.
Exclusively, those who have reported problems with dwi2response have been
using data with a low b-value and a large number of directions. This poses
two problems. Firstly, a large number of directions will result in
automatic selection of a large *lmax*, which makes both the initial SH
transformation and CSD considerably more computationally expensive.
Secondly, at low b-values, there is simply *no power present* in these
higher harmonic orders; the magnitude of diffusion sensitisation is not
adequate to introduce such high-frequency components into the measured
signal profile. I would highly recommend reading this article
<http://onlinelibrary.wiley.com/doi/10.1002/nbm.3017/abstract> to see the
evidence of why this is the case.

So, for anyone using the dwi2response command, I have the following
recommendations:

   - Update your MRtrix3 install using "git pull; ./build"; this will
   change a couple of the default parameters in dwi2response and hopefully
   prevent very long execution times.
   - Use the -mask option! You're not going to use any voxels outside the
   brain to determine the single-fibre response function, so no point in
   processing them.
   - Set the -lmax option manually based on *both *the number of directions
   in the b-value shell you are using* and* the b-value of that shell.
   Particularly if you get the "ERROR: matrix must be positive definite"
   error: this means your *lmax* is *waaay* too high.

Normally I'd say "happy tracking", but I guess in this case it should be
"happy RF estimation"!

Rob


--

*Robert Smith, Ph.D*
Research Officer, Imaging Division

The Florey Institute of Neuroscience and Mental Health
Melbourne Brain Centre - Austin Campus
245 Burgundy Street
Heidelberg Vic 3084
Ph: +61 3 9035 7128
Fax: +61 3 9035 7301
www.florey.edu.au


On Tue, Jun 17, 2014 at 3:15 AM, Donald Tournier <jdtournier at gmail.com>
wrote:

>  Hi Romain,
>
> Yes, we've had reports of people having trouble with the new dwi2response
> app. This is something we're going to be looking into in the near future...
>
> Cheers,
> Donald
>
> --
> Dr J-Donald Tournier (PhD)
>
> Senior Lecturer, Biomedical Engineering
> Division of Imaging Sciences & Biomedical Engineering
> King's College London
>
> A: Department of Perinatal Imaging & Health, 1st Floor South Wing, St
> Thomas' Hospital, London. SE1 7EH
> T: +44 (0)20 7188 7118 ext 53613
> W:
> http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering
>
> On 16 Jun 2014 16:49, "romain valabregue" <romain.valabregue at upmc.fr>
> wrote:
>
>> Hello
>>
>> I get an error on several dataset with de dwi2response :
>>
>>  dwi2response -grad grad.b -mask mask_erode3.nii  -sf sf_out.nii
>> rfsl_eddycor.nii resp.txt
>>
>> dwi2response: initialising response function...  100%
>> dwi2response: optimising response function..   .   gsl:
>> lib/math/cholesky.cpp:151: ERROR: matrix must be positive definite
>> Default GSL error handler invoked.
>> Abandon
>>
>> It takes quite a long time before getting the error (about 1/2 an to 1
>> hour)
>>
>>
>> Although it works ok with connectome B3000 data, I have problem with
>> other data set that was working before with mrtrix2
>>
>> Many thanks
>>
>> Romain
>>
>> _______________________________________________
>> Mrtrix-discussion mailing list
>> Mrtrix-discussion at www.nitrc.org
>> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20140619/ca3e4748/attachment.html>