[MINC-development] -O3 breaks N3

[Claude LEPAGE]

>
> Well, there are no tests to convince one that the current
> implimentatation is not broken either. I remember an experiment that
> Claude performed a couple of years ago , which involved improving N3 .
> Unfortunately it completely changed the results of cortical thickness
> analysis that Alan's group was doing at the time.

Just as a recall to everyone, my change was a bug fix to divide
the damping term by N, where N is the number of degrees of freedom.
This N was missing in the initial implementation. The fix occurred
starting at version 1.11, so reproducing results from earlier 
versions will be meaningless. It is also important to note that
the versions prior to 1.11 are not totally wrong. The damping term
was calibrated in such a way to produce good results in MNI
stereotaxic space with 1mm isotropic voxels. The old N3 (1.10)
was indeed wrong with non-isotropic voxels other than 1mm (many
scanners can now produce scans at 0.5mm). The new N3 is voxel-size
invariant, as the formulation states in the original paper by Sled.

As always in our field, it is rather easier to invalidate software
than to validate it. 

Claude