[Insight-developers] MutualInformation / Registration Validation

[Luis Ibanez]

Hi,

I just got some input from Guido Gerig about the software
that is used here for registration.

This is called MIRIT and was developed at the University
of Leuven. Dick Vandermeulen seems to be the person that
could be contacted for getting access to a binary version.

This package not only does MutualInformation, it is also
possible to select the metric to apply, it follows a multi-
scale approach also, and has particular way for implementing
the optimization.

--

MIRIT was one of the techniques used on the "Retrospective
Registration Evaluation Project" performed by J.M.
Fitzpatrck at Vanderbilt:

     http://www.vuse.vanderbilt.edu/~jayw


Results from this validation project were published on:

J.West, J.M.Fitzpatric ... et al (it is a long list
of authors because it includes all those who provided
code and submited registrations)
"Comparision and Evaluation of Retrospective intermodality
 brain image registration techniques",
J. Computer Assist. Tomogr. Vol 21. pp 554-566, 1997.

The PDF/Postcript version can be downloaded from:

http://www.vuse.vanderbilt.edu/~jayw/online_files/jcat_97.pdf


----


In any way, it looks like the same validation problem
(Just to know if our implementation of somebody's algorithm
is correct) will exist with any other method or technique,
and probably contacting the authors is the reasonable way
out, even if this doesn't seems to be the fastest one.

Probably we should make an initial try on a Validation
Dashboard, at least taking registration as a prototype,
and then, according to the "extreme programing" paradigm,
improve it progresively. If developers can reproduce the
tests reported by users, it will be much easier to figure
out whether problems arise a cause of: wrong parameters
settings, bugs on the code, or just because the method
is not appropriated to the particular problem.



Luis