[Mrtrix-discussion] hi

Sun Aug 31 16:33:54 PDT 2008

Hi Vinod,

Some interesting questions you have here! Here's my 2 cents, for what
it's worth:

(1) Although the docs say that the first coefficient should be similar
to the b=0 signal in white matter, this should not be taken too
literally. The reason this statement was included in the first place was
to alert the user to cases when the estimation of the SH coefficients is
badly determined. This can happen when you have for example a data set
with the same directions repeated twice: in this case, MRtrix will
mistakenly assume that all directions are unique, and attempt to use a
higher lmax than can be supported by the data. In such cases, the
estimated response is complete garbage, and the coefficients will be off
by many orders of magnitude. So, as long as the first coefficient is in
right ball park, you should be ok. Looks to me like you first response
coefficients were pretty good...

In fact, the first SH coefficient is only indirectly related to the b=0
signal, as it depends only on the DW signal. The first SH coefficient
should be equal to sqrt(4*pi)*(mean DW signal). Since sqrt(4*pi) ~= 4,
and the mean DW signal is approximately one quarter of the b=0 signal,
it just so happens that the first SH coefficient will be similar to the
b=0 signal. This is coincidental rather than a hard relationship though,
so should not be taken too literally (it might not hold at lower
b-values for example).

What is important though, and you do allude to it, is to make sure that
the single-fibre mask is good. The best thing to do is to overlay it on
the FA map using the ROI tool. As long as the voxels in the mask are in
nicely coherent white matter regions, you should be ok. And also bear in
mind that 200 voxels distributed over a whole brain data set doesn't
look like much at all - it looks like there are far fewer voxels than
there actually are. I don't think you would ever need to lower the
threshold lower than 0.5 - if you do, there might something odd going
on.

(2) I don't have a sample data set at the moment, but that is a good
idea. I might produce one if I ever have the time...

(3) I have just added NIfTI support to MRtrix, for precisely the same
reasons as you: interoperability with other packages such as SPM5 & FSL.
A few people want to coregister their anatomy with the DWI, and would
need to use SPM or FSL for that. This will be included in the next
release, which will be imminent (I just need to write some documentation
and upload).

However, I would not recommend producing Analyse images to import into
MRtrix. For whatever reasons, it is (or has become) a poorly defined
format, with ambiguous left-right ordering. Moreover, the information
that can be stored in the header is limited. In particular, there is no
transform information to orient the slices, and no scope to include the
DW orientation scheme within it. As a result, if you try to import your
DW images in Analyse format, and they happened to have be acquired
tilted (not straight axial), you might have a mismatch between the DW
gradient orientations you supply and the images themselves. Directly
importing NIfTI images will be a much safer option - please don't use
Analyse if you can avoid it.

Hope that help. I'll go work on the next release now...

Cheers,

Donald.

On Fri, 2008-08-29 at 05:20 -0700, vinod kumar wrote:
> Hello to all mrtrix users and developers :) 
> 
> I am new to mrtrix and trying to tune some parameters for my data.
> here my very basic questions. 
> 
> (1). 
> This is suggested in the mrtrix docs '''' The first value should be
> positive, and have an amplitude similar to the b=0 brain white matter
> signal (you can verify this by loading the dwi.mif image and placing
> the focus within a white matter region - the intensity is displaced in
> the bottom right of the statusbar )''' 
> 
> So, when I go to white matter than I found some places 146, 157, 99,
> 148 and it varies around 100 to 180, just I checked on few places. 
> 
> So, when I estimated response for 126 direction dataset, using -lmax
> 10, 12, 14, I found very minor change between all the resulted
> response.txt. and the first value of response.txt was higher than the
> white matter value of my data. 
> 
> >estimate_response dwi.mif sf_th0_6.mif -lmax 10 response_sf_th0_6.txt
> -grad encoding.b 
> 
> response text file looks like this. 
> 172.802 -82.9005 38.419 -12.4871 2.83802 -0.640453 
> 
> 172 is bit higher with the white matter values of my data, so I tried
> with lmax 14 and lower and looked at the respone text, for example
> with lmax 14 it has minor change. 
> 
> 172.655 -83.0056 38.5118 -12.6871 2.58813 -0.679325 0.196577
> -0.126814 
> 
> 
> So, I looked up and checked the sf.mif ( Mask of single-fibre
> voxels ), and I found it has few voxels only. (also in the site
> written a mask containing a few hundred voxels ). I created it at FA =
> 0.6 threshold.  
> see attached with the email (sf_th0_6.mif). 
> 
> than I lowered down the threshold to 0.4 (sf_th0_4.mif), and estimated
> response again (lmax10)
> 159.062 -64.3264 26.9399 -8.18025 1.79083 -0.315482 
> 
> sf_th0_2.mif (0.2 threshold) + lmax10 > response.txt
> 131.734 -36.8685 13.1795 -3.64864 0.777517 -0.140337 
> 
> 
> sf_th0_1.mif ( 0.1 threshold ) + lmax12 > response.txt
> 124.456 -26.1018 8.307 -2.14918 0.443511 -0.0872688 
> 
> so, 124.456 was nearer than other response.txt, So, I took this one
> and performed CSD with lmax12
> now when I overly odf, than three pictures have some thing else than
> normal (attached with email). 
> 
> I think this was not a good idea to go to sf.mif threshold 0.1. 
> 
> So the question is how I can get the best response.txt when data has
> different white matter values, and the output of response estimate is
> higher than the lowest value of white matter ?. 
> 
> 
> (2). Is there any sample dataset, which fits on the help file standard
> parametes, so that one could get an idea, how the output should look
> like for his data also. 
> 
> (3). I preprocess data using FSL (eddy current correction) and after
> that I took them out in .nii format and used LONI debeller for analyze
> conversion. and later on used mrconvert, and converted in .mif.  
> I think this is okay, if some one have any better way for importing
> nifti data, let me know.
> 
> thanx 
> vinod
> 
> 
> 
> _______________________________________________
> 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 9496 4078