Hello Simon,
hello list
we are trying to adopt the Anatomy toolbox for an automated volumetry
approach, combining VBM preprocessing with the probability maps.
I am interested in any opinions on this - may be a somebody has also used a
mask & VBM approach for regional volumetry.
Images have been normalised to MNI-305-space and later to a group templates,
before (optimized) segmentation.
The basic ideas is to use the probabilty maps and multiply them on (modulated)
grey and white matter maps. So, by using the probability information of the
cytoarch. we perform a weighting of the total volume of the region, assuming
that collecting volume from the 'core region' is more likely to belong to the
selected region than more distant voxels.
Questions:
1. As the spatial variance is already encoded in the probabilty maps, it seems
that smoothing of the modulated maps is not necessary any more. Is there a
rationale to still do so? (Other people using binary masks have at least
reported using smoothed maps).
2. The hightest theoretical value is 1 (or 255) - however, some regions do not
reach this value due do the high anatomical variability. Do we need some
rescaling of the maps? (Or rather keep the original values to maintain the
different regions comparable?)
3. There is of course considerable overlap between the regions. We were
debating if as an alternative the binary maps should be used?
4. Could the result from subregions be added to a total volume? Or should we
rather add (or fuse) the P maps before?
5. What do you think about the difficult regions that do not offer a good
contrast between grey and white matter on anatomical MR scans, as e. g. the
amygdala. If grey and white cannot be separated, it seems the cytoarch.maps
are not suited for such a volumetry. Is this correct?
Thanks a lot in advance,
Philipp
|