Hi,
On 27 Aug 2009, at 13:26, Mark Rees wrote:
> Hi,
>
> I have a large single, large cluster (1000+ voxels) present after TFCE
> correction at 0.001. To explore this further I want to run some
> tractography from it. As the cluster is so large tractography just
> using
> this cluster is difficult to interpret, and selecting the COG or
> voxel with
> peak p/t value within this cluster would lose a lot of information.
>
> In order to break down the cluster into smaller clusters and seed
> from them
> I was wondering which of the following would be most valid or
> whether they
> would give the same result:
>
> Approach 1:
> a) Take corrp image (image A) and apply clustertool using parameter
> settings
> that break it down as much as possible (eg high threshold and low
> connectivity value)
I don't know what clustertool is? However yes in general I would not
change the connectivity in clustering for this purpose, just change
the threshold, to allow you to find the most significant sub-parts of
the large cluster. You may find that there isn't enough value-
resolution in the corrp image to allow this to be done very usefully,
in which case you can instead re-run randomise, turning on the TFCE-
enhanced tstat output option and setting the number of permutations to
1. You can then apply the re-clustering to the TFCE-enhanced image.
> b) Mask zmax voxels in each cluster onto original corrp image (image
> B) --
> ie left with an image of zmax voxels for each cluster
that's good too, or you could seed from the whole sub-clusters if you
think they're small enough
> c) Deproject and backnormalise image B
> d) Create a small sphere aournd the voxels in image B and use as seeds
>
> Am I right in thinking this might not work as step c) may not result
> in
> single voxels in native space for each cluster? (ie in deproject the
> no of
> of voxels in the input does not equal the no in the output)
hopefully the latest versions should in general backwarp a single
point to at least a single point - hopefully.
>
> Approach 2:
> As above but create sphere before deprojection and back
> normalisation -
> might this have problems if the sphere lies off the group skeleton ?
Yes - that won't work, for that reason
>
> Approach 3:
> a) Take corrp image (image A) and deproject into each subject
> b) Then apply clustertool with the same threshold and connectivity
> parameters to each subject
> c) Within each subject only accept clusters over certain size eg 10
> voxels
> d) Within these clusters take zmax voxel and create sphere of set
> radius
> around them
> e) Now backnormalise all clusters into native space and use as seed
> in each
> subject
That's ok - but I think Approach 1 is probably best.
Cheers.
>
> Or perhaps someone might have a much better suggestion ?
>
> Thanks.
>
> Mark
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|