Hi all.
There's a short version and a longer one, for those bored or wanting to read
a bit more:
Short:
Is tbss viable for intra-individual comparisons (i.e. paired testing ny e.g.
different time points), minus the statistics in case of only very small n's?
Long:
I didn't find a posting on the subject here and there's not much literature
(as one says: to the author's knowledge...). Take the following situation:
there are some neurodegenerative diseases with rather quick progression and
they're rare, without much overall structural change; so your looking at a
small n of say 2 or 3 in a few of years with a pair of intra-individual
DTI's being acquired. It's not sensible to carry out group-wise testing, so
all you want to see are differences in patients over time. Moreover, you
expect change not only in well defined tracts as such, but also diffuse in
areas like the thalamus. <- this is theoretical and is interchangeable with
e.g. MS or any other intra-individual change situation.
So here are the methodological questions:
1.) I think identifying a common registration target such as FMRIB58_FA is
sensible, but is it really? If in fact I'm comparing the subject at t1 and
t2 it may make more sense to register t1 on t2 of vice-versa (i.e.
tbss_2_reg ... -t). Is there any empirical evidence which approach is better
in light of sparse data?
2.) One could stop here, do a bit of smoothing and subtract t2 from t1, or
do other descriptives. But should one stop here or skeletonize? If yes,
again the question crops up: use the patients FA to create a mean or use the
standard skeleton?
3.) If one does skeletonize, then subtracting the all_FA_skeletonized
volumes should result in visualizing temporal differences between t1 and t2,
or does it?
4.) As there is a distance function mapping FA (or other things) onto the
skeleton, what if one is in fact interested in things more distant from the
skeletonized tracts, e.g. thalamus?
Regards.
Stefan Kreisel
|