Dear David,
>
> I am wondering whether when using fnirt it is possible to use successively higher isotropic resolution of knot-spacing at each iteration level, as in --warpres=8,6,4,2
>
> Is it possible that fnirt might support something like this now or in the future?
it is supported through the --inwarp mechanism. What you do is that you run fnirt to one resolution and if you want to increase the warp resolution you launch fnirt again, this time with a higher requested resolution and passing the warps from the previous fnirt as the --inwarp parameter. Note that as you go to higher resolutions the Gauss-Newton optimisation that is default in fnirt will become very memory hungry and if your computer can't handle that you might want to switch to --minmet=scg.
It is a little silly it is like that, but the reason is that fnirt allows for different warp resolution along the different axes (which was another silly initial design choice) which means that --warpres=8,6,4 would simply be interpreted as 8mm along x, 6mm along y etc.
>
> Also, I am wondering if it is possible to have fnirt spit out the final correlation coefficient computed between the two images within the ref mask.
fnirt doesn't use correlation and doesn't calculate it. I _think_ it would be easy enough to do with flirt (after fnirt has finished). Have a look at the flirt options.
Jesper
>
> Thank you very much,
>
> David
|