Thanks Gordon and Jesper,
I share Gordon's view because this is my trouble as well:
Analysing eigenvalues (and/or lamba parallel and perpendicular
or RA) of small, noisy ROI gives problems with li<0.
But more important are the analysis in the form of normalized histograms
of ROIs, VOIs (of eigenvalues and invariants!) that can skew the analysis
if one does not carefully eliminate negative eigenvalues and their influence
on invariants beforehand. Or would you suggest a better strategy?
Thanks everyone, Boogie
|