[Insight-users] ImageRegistration4 example problems

[Paul Hughett]

> Mutual Information, on the other hand, will not behave well in
> same-modality registration. Altough in theory it could do it, in
> practice the joint-probability distribution of gray levels is
> concentrated into a small region of the joint histogram (the
> diagonal...) making that the estimations mutual information to be
> poorly evaluated.

That's not my experience.  I've been doing MI registration of
T1-weighted MRIs and gotten consistently good performance, both within
subject and between subjects; it seems to work about as well as
correlation or mean-square error.  It may or may not be relevant that
I'm using my own toolkit rather than Insight.

Paul Hughett