Hi there Jon,
yes there is a way of specifying sets of matching points in the ---ref and --in scans so that registration is driven both by aligning these points and by the "usual" image intensity metric. It's a hidden feature as we haven't found it to be particularly useful and since I fear it would generate too much support work.
The difficulty with it is to find points that have sufficient curvature in 3 dimensions that a point can be reliably identified. Most landmarks in the brain are surfaces rather than points, and we have no mechanism for including (or generating) surfaces. I would have thought that the same was true for the spinal cord (even more so maybe). Imagine trying to finding matching landmarks on two sausages. You'd struggle to find anything but the ends.
If you think you can find reliable landmarks you can email me off list and I'll show you how to use them.
Jesper
On 21 May 2013, at 12:03, Jonathan Brooks wrote:
> Hi Jesper,
>
> I have a question about fine control in FNIRT. I seem to recall that there was a way to specify a one to one matching between points on the input and reference volume when running FNIRT. Did I imagine this!?
>
> The reason I ask is because (for some) the anatomy I'm working with can't be brought into sufficient alignment with the reference using affine registration alone (as I have to deal with curvature of the neck).
>
> If you can suggest any other approaches that might be useful it would be much appreciated.
>
> Cheers, Jon
>
> --
> Jonathan Brooks (PhD)
> CRiCBristol Senior Research Fellow
> University of Bristol, St Michael's Hill, Bristol, UK
> T: +44(0)117-342-1512 W: http://www.cricbristol.ac.uk
>
|