I am using the HDW toolbox to warp images from multiple timepoints together
for TBM. These are scans of patients with neurodegenerative diseases so
there is considerable progression of atrophy. I have two questions...
(1) HDW does a decent job of matching scans to one another, but I'm
wondering if it could be improved by tweaking the parameters. In the
attached figure, note the enlargement of the ventricle under the crosshairs
in year 4 (middle image) as compared to year 1 (left image). When year 4 is
warped to year 1 (rightmost image), the ventricle is reduced in size as it
should be, but not quite back to the year 1 size. I think this implies that
the Jacobians will fail to capture the full extent of the atrophy.
I tried changing the two available parameters. Reducing the regularization
parameter helped a bit, with the best performance in this case at about 1
(as opposed to the default 4), but messy results elsewhere in the brain.
However I got odd results when changing the number of iterations. Whereas
fewer iterations gave poorer results, 12 iterations also gave poorer results
(less warping) than the default 8, and 16 iterations actually gave identical
results to the default 8 (and by identical, I mean that the command diff
revealed no differences in the files). That seems strange.
Any thoughts on how to tweak the parameters beyond what I have tried?
(2) Looking at the Jacobian images, ventricular enlargement, which occurs in
all subjects, is usually accompanied by a "halo" of contraction around the
ventricles (see second attachment). I suspect that we should not be so
confident in localizing the contraction to the regions immediately around
the ventricles; it should probably be distributed throughout white matter.
Any thoughts on this, and what to do about it?
Thanks very much for anyone's thoughts on these questions.