Hi,
> I have some queries about FIRST. BTW this seems to be a great tool,
> which
> makes it very practical to analyze subcortical volume/shape, but
> could also
> be very useful for providing individual anatomical ROIs for functional
> studies.
Thanks.
> I have been using this for an exploratory investigation hippocampal
> volume/shape in some subjects from a VBM (SPM) study I have already
> run.
>
> I followed the instructions on the first webpage
> ("http://www.fmrib.ox.ac.uk/fsl/first/index.html"), and was able to
> segment
> the hippocampus and subcortical structures. I then used first-utils
> to run a
> vertex analysis on the results, again following the instructions
> closely,
> except mine is a correlation design (rather than group comparison).
> I was
> able to view the results in FSL-view, where they corresponded well
> with what
> I would have expected from the VBM analysis. I think they add to the
> VBM
> results, and I'd like to include them in any publication, but before
> that
> I'd like to read a more detailed description of the vertex analysis
> (as
> opposed to the segmentation).
>
> Here are some concrete questions:
>
> i) For each vertex, what exactly is being fitted to the regressors
> in the
> GLM?
The GLM is formulated in a multi-variate way, testing the coordinate
locations
(which a 3-vectors) for that vertex over the different individuals
versus the regressor
of interest. Often the regressor of interest is the group difference
regressor, but it
can be a non-categorical one in which case you get the correlation,
like you want.
Note that everything is demeaned in this GLM. Also, if you specify
several different
regressors, they are actually fitted with separate GLMs and results
given for each
separately.
> ii) Results: I'd also like to know (daft question) what the units
> are in the
> vtk-files that are produced as output (Z-scores?) and displayed in
> FSL, and
> more generally what is the format of the output files. It might also
> be
> useful to know the bvars file format.
The initial output gives F statistic values. You can convert this to
p values
using the script surface_fdr, which is very simple and will show you
some
ways to manipulate these files.
The vtk file format is pretty simple: all text and start with the
vertex coordinates,
then the connectivity (in triangles), then any scalars or vectors that
are attached
to the vertices. Note that the list of scalar or vector values is
done in the same
order as the vertices are initially listed.
The bvars file is much more complex, so I won't describe it here. It
relates to
the shape and intensity model on which FIRST is founded.
> I found it difficult to control the visualisation in FSL-view. So
> far as I
> can see, it is hard to include more than one mesh (you can add them,
> but you
> cannot select them individually to vary e.g., the range of the
> colour bar).
> For example, I'd like to have more control over the colourmap used,
> and I'd
> like a more constrained view (e.g., axial, orthographic projection
> avoiding
> perspective artefacts) and to include both one hippocampus. For these
> reasons I'd prefer to import the data into e.g., matlab, blender in
> order to
> produce better figures. The format of the vtk files seems very
> readable, but
> to avoid mistakes it would be useful to get an authoritative
> description of
> the file format.
Yes, FSLView is currently limited in what it can do with meshes.
There should be easy ways to read in the appropriate information to
MATLAB
and do your visualisation there. I haven't personally read in a vtk-
file to MATLAB
but I suspect that there are tools to do this. If not, you can fairly
easily divide
the file into the appropriate chunks and read them in separately.
> I understand that two meshes are produced (one representing the mean
> shape
> of subjects where the covariate is less than the mean, one the mean
> for
> those for whom it is greater - one can morph between the two using
> the mesh
> slider. However, this doesn't capture the full range of variation in
> the
> data, and it might also be interesting to see the vertex-wise results
> projected on the more extreme meshes or, perhaps, on a "standard"
> mesh.
As the order of the vertices is the same for all files in a given
shape analysis,
you can simply load in the coordinates of each of the native image
meshes and
average them (preferrably after alignment to a common pose or space).
For
instance, you can apply the transformation to standard space and then
average
the meshes in standard space. This just requires a matrix
multiplication of the
flirt matrix and the coordinates (as these are in appropriate units).
You may not
get shifted coordinates back, but for visualisation purposes they will
do fine.
All the best,
Mark
|