[Mrtrix-discussion] Mrstats question

[Donald Tournier]

Hi Dan,

This is probably a little late, but I just realised your question had gone
unanswered.

Firstly, do the min and max results refer to the lowest and highest FA
> values in the voxel mask I'm running the function on?
>

Yes, they do.


> The function consistently reports min values well below the FA cutoff i've
> used in generating my tracts, often below 0.05.
>

I'm guessing you're generating a track count/probability map using
tracks2prob, and using that as the mask for mrstats? Although getting values
lower than your cutoff is probably unexpected, remember that the tracking is
done using trilinear interpolation. It's likely that tracks will run into
low anisotropy voxels (lower anisotropy than your cutoff) if the adjacent
voxels have high anisotropy, since in this case the contribution from the
high anisotropy voxels will increase the anisotropy at the interpolated
position. Hope that makes sense.

By the way, the explanation above is valid only for the DT_STREAM method.
For both SD_ methods, the cutoff actually refers to the FOD amplitude in the
direction of tracking - FA is NEVER considered for these tracking methods.
If your tracking goes through crossing fibre voxels with low anisotropy but
well-defined orientations, that will indeed also be reflected in your
mrstats results.


> Secondly, the max value is occasionally >1.0. Looking at the FA maps I'm
> generating on mrtrixs I occasionally see voxels with FA values >1.0. I think
> this may be related to noise as the voxels are in regions where diffusion is
> very restricted in one direction. Is there anyway of correcting for this?
>

Yes, this problem does happen all too frequently. It is, as you say, due to
noise in highly restricted diffusion regions/orientations. There are methods
to deal with it using constrained non-linear fitting methods (I can't think
of any references off the top of my head), but the vast majority of
implementations (including the one in MRtrix) use the simple log-linear
least-squares fit, which has been shown to be sub-optimal in numerous
publications...

I may or may not include improved fitting algorithms in future releases of
MRtrix, but given how little spare time I have these days, it's unlikely to
happen any time soon...

Cheers,

Donald




Thanks in advance for your advise
>
>
> Dan Lumsden
> Research Fellow
> Complex Motor Disorder Service
> Evelina Children's Hospital
> London
>
> _______________________________________________
> 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/20110520/93760853/attachment.html