Hello,
I applied permutation testing with the randomise script on siena flow
images, and here a couple of questions arised:
Images were processed as recommended with the siena_flow2tal script,
smoothed (10 mm) and merged.
(1) The pairs of original image analysed vary in the time delta between
them, i. e. some were obtained 12 months, others 15 months apart. I would
like to take this into the model in the way that scans longer apart should
be less weighted as their atrophy pattern is 'falsely strong'. Could this be
done by a regressor 12/delta with delta being the months between the scans?
(Apart from the of course incorrect assumption that the atrophy process is
progressing linearly with time)
If it is working with such a regressor, how is it dealt technically for
randomise - orthogonalize it and feed it in with -x?
(2) Is variance smoothing recommendable when using images from SIENA?
(3) Is more smoothing than 10 mm in the preprocessing appropriate with more
deformated original brains (Alzheimer) or is to your experience the
normalization of the flow-images sufficient?
Thanks in advance,
Philipp Saemann
Max-Planck-Institute of Psychiatry
Kraepelinstr 2-10
80804 Munich
[log in to unmask]
|