Hi Emma
The curvature threshold used by default in probtrackx is only there to
avoid paths tracking back on themselves and artificially increasing
the probability along regions that were already visited. The angle of
80 degrees is high enough, however, to allow the sampling to account
for local uncertainty. We see the spread in the path spatial
distribution not as increased false positives, but as increased
uncertainty :)
That said [or written], I have tried the thalamus segmentation using
decreasing threshold angles (increasing curvature thresholds) - see
attachment.
The picture shows that the general pattern is conserved for 60
degrees, which is a reasonable threshold, but voxels "inside" the
thalamus tend to have higher directional uncertainty, which leads to
the pattern you get for the conservative threshold of 45 degrees. Note
that even at this level, the general pattern is conserved, especially
on the external shell of the thalamus.
Cheers,
Saad.
On 29 Jan 2009, at 12:04, Emma Robinson wrote:
> Hi
>
> Can someone please explain to me the motivation for having a curvature
> threshold of 80 degrees. It seems quite high to me however I have
> done some
> experiments on adult data where it is clear that, using only 15
> gradients
> directions, it is not possible to get tracts that you would expect
> to find
> such as those between the thalamus and motor/sensory cortex with a
> lower
> threshold. Nevertheless might it not also generate a number of false
> positive connections?
>
> Thanks
>
> Emma
>
Saad Jbabdi
Oxford University FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
+44 (0) 1865 222545 (fax 717)
www.fmrib.ox.ac.uk/~saad
|