Hi,
Correlation Ratio is very different from Normalised Correlation.
The correlation ratio is designed to work with different modalities
where you can define a general non-linear function that relates the
intensities in one image to the intensities in the other.
We have found that this works better than mutual information in
general for T1 to EPI (possibly because it is a little more constrained
than Mutual Information which doesn't even assume that the same
tissues can be seen in each image) and so this is our default both in
Melodic and in our functional analysis tool, FEAT.
I think you'll find that this default should work perfectly well for
you.
All the best,
Mark
On 15 May 2009, at 11:53, Stefan Roosendaal wrote:
> Dear FSL people,
>
> I recently upgraded to FSL 4.1.3, and was happy to see that non-linear
> registration can be used in Melodic for the highres image to
> standard space.
>
> My question, however, is about the first part of the registration,
> example_func to highres. It seems to use by default corratio as a cost
> function, where I would think that mutual info as a cost function is
> more
> appropriate, since two contrasts (EPI to T1) are involved. It can
> not be
> changed in the Melodic GUI.
> What are you're thoughts on this?
>
> best regards,
> Stefan Roosendaal
> Amsterdam.
>
|