[Mrtrix-discussion] csdeconv: failed to converge

[Donald Tournier]

Hi Markus,

csdeconv outputs one 'failure to converge' message per voxel where the
maximum number of CSD iterations was exceeded (by default, the maximum is
50 iterations). So unless you're getting thousands of these messages, the
proportion of affected voxels is probably fairly low. Moreover, these
issues typically arise in GM or CSF, without affecting WM voxels. Finally,
even in affected voxels, this is usually not a problem, as what generally
happens is that the algorithm oscillates between two equally good
solutions. If you allow the program to run to completion and inspect the
resulting FODs in MRView (orientation plot sidebar with overlay), my guess
is the results will look OK.

You'll get bigger problems if you get "not enough negative directions"
messages. In these cases, the affected voxels are assigned a NaN value, and
so can't be tracked through. Note this only happens when using
super-resolution (i.e. lmax is greater than the data can natively support).

Hope that helps.
Cheers,

Donald.


On 20 December 2011 09:34, Markus Gschwind <markus.gschwind at gmail.com>wrote:

> Dear experts,
>
> I'd like to ask what could be the reasons whay csdeconv fails and gives an
> error like
>
> > csdeconv data.mif response.txt -mask mask.nii -grad gradtable.txt
> CSDm.mif
> csdeconv: performing constrained spherical deconvolution...   0%csdeconv:
> failed to converge
> csdeconv: failed to converge
> csdeconv: failed to converge
> csdeconv: failed to converge
>  csdeconv: failed to converge
> csdeconv: failed to converge
> csdeconv: failed to converge
> etc.
>
> I tried with different -lmax and with different thresholded sf.mif (values
> 0.7, 0.6, 0.5, 0.4).
>
> Principally the commands are correct as they worked with other brains, but
> not with the one in question ( a lesioned one does it not work with brain
> lesions?).
> How should I proceed best  in order to debug?
>
> Thanks in advance!
> Markus
>
> _______________________________________________
> Mrtrix-discussion mailing list
> Mrtrix-discussion at www.nitrc.org
> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>
>


-- 
Jacques-Donald Tournier (PhD)
Brain Research Institute, Melbourne, Australia
Tel: +61 (0)3 9035 7033
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20111220/16f48f0a/attachment.html