[MINC-users] 2D-3D registration

[Andrew Janke]

Hi Laurence,

On Fri, May 29, 2009 at 00:48, Laurence MERCIER
<laurence at bic.mni.mcgill.ca> wrote:
> I have a bunch of 2D minc images (ultrasound) that I would like to
> register to a 3D minc volume (MR). I can obviously reconstruct a 3D
> volumes from my ultrasound slices and then register volume to volume, but
> I would like to know if there is a way to directly register the 2D images.

Well no-one else seems to have replied to this yet so I will take a
hack. I was hoping that the likes of Mallar might have something to
add in here..

I do know that direct registration of 2D to 3D using minctracc can be
somewhat problematic as the rotations out of plane tend to go haywire
very fast. That said there are a few little hacks I have had success
with in the past (no doubt louis has some too).

1. Expand the data in the out of plane direction before registration.
If you have a single Z slice, duplicate the slice (mincconcat can do
this) about 10 times first and then attempt registration, if you
slowly blur the slices out towards the edge things also sometimes work
better.  ie:

top - multiply center by 0.1
next - multiply by 0.2
...
Center (original) slice - multiply bu 1.0
...
next - multiply by 0.2
bottom - multiply center by 0.1

2. reduce the w_rotations parameter for the out of plane rotation
directions to ensure that only small out of place rotations are
attempted. Also be sure to start with -identity and -lsq9  to start
with as the PAT will definitely go awry when it aligns the COG's of
the two. Note that this will only work if you have a fairly good idea
of where the slice is to start with.  Perhaps there are some
heuristics in the ultrasound slice that you can take advantage of.

have fun.


--
Andrew Janke
(a.janke at gmail.com || http://a.janke.googlepages.com/)
Canberra->Australia    +61 (402) 700 883