[Insight-users] [Insight-registration] Registration of one sub-brain to full-brain

[Torsten Rohlfing]

  Hi Rupert.

On 10/23/2010 10:25 AM, Rupert Brooks wrote:
> I'm curious about your statement that partial data does not register
> well in the non-rigid case.  I'm not sure i would agree that its true
> in all cases, i've definitely used masks in non-rigid registration to
> good effect.  I would think that a half brain, properly masked, would
> match well with a full brain.  I didn't try though, so i might be
> wrong.  In any case, i'd be interested to hear some specifics.
Well, I wasn't saying it NEVER works, but it is nowhere near as 
successful as affine registration of partial data (again, based on my 
own subjective experience). The trouble with nonrigid when using an 
image mask is that the deformation can selectively move things in and 
out of the masked area in one place without being bound by any 
constraints imposed by other places.

As an example, when I try to register a skull-stripped brain to a 
whole-head image, then there is only one object that "looks" like a 
brain in the complete image, so masked registration works pretty well 
finding that as the registration outcome.

With nonrigid in the same example, there may well be parts of the skull 
in some places that fit the brain in that area of the stripped image 
better than the neighbouring brain, so the nonrigid transformation may 
pull these regions into the wrong place.

Hope this makes sense. Again, this is just my empirical experience, no 
hard science ;)

Best,
   Torsten
-------------- next part --------------
A non-text attachment was scrubbed...
Name: torsten.vcf
Type: text/x-vcard
Size: 373 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20101023/6a9dc510/attachment.vcf>