[Insight-users] 3 image registration

[Torsten Rohlfing]

Niels:

This is a bit of a shot in the dark since I am not really using the ITK 
registration classes myself, but as a first "hack" you could try 
optimizing the sum of the mutual information measures computed 
independently for US-MR1 and US-MR2. My guess (again, shot, dark, etc.) 
is that you may have to construct some sort of wrapper class that 
pretends to the optimizer that it is a simple image-to-image similarity 
measure, but internally computes two (or more) similarities and sums them.

As for a more general, true multi-dimensional MI, that's very much open 
research, and people still debate what the best way is to formulate MI 
(or rather: a good registration metric analogous to MI) for more than 
two images. Then, there is the additional implementation issue for 
entropy computation, which is that higher-dimensional histograms are 
increasingly difficult to populate with samples, and computation of the 
necessary entropies gets complicated. There are ways to use vector hash 
tables instead of actual histograms for representation, or minimal 
spanning tree algorithms for entropy computation in high-dimensional 
spaces, but to the best of my knowledge, none of this is currently in 
ITK. (I'd gladly be corrected here by someone else who may know better 
of course). Also, I have yet to see any of these actually working 
reliably and efficiently.

Best,
  Torsten

> Hello itk-users,
>
> I would really appreciate some help in solving the following problem:
> I'm trying to solve a multiple image registration problem and wondered 
> if the ITK already contained some components for this.
> To be more precise, I am trying to register a (fixed) 3D ultrasound 
> image to a couple of (moving) 3D MR images. The MR images have the same 
> domain, so it's not a general 3 image registration problem, but more 
> like a scalar-image-to-double-valued -'vector'-image-problem. I do not 
> know how to adapt the 'itkMattesMutualInformationMetric' (I prefer to 
> use) in order to accept two moving images or a moving vector image 
> (probably due to the fact that the neccesary 3D histogram construction 
> is rather tricky to implement in a meaningful way??)
> Does anyone know how to achieve this or have some hints on how to proceed?
>
> Thanks in advance
>
> Niels
>
>
>   

-- 
Torsten Rohlfing, PhD          SRI International, Neuroscience Program
 Research Scientist             333 Ravenswood Ave, Menlo Park, CA 94025
  Phone: ++1 (650) 859-3379      Fax: ++1 (650) 859-2743
   torsten at synapse.sri.com        http://www.stanford.edu/~rohlfing/

     "Though this be madness, yet there is a method in't"

-------------- next part --------------
A non-text attachment was scrubbed...
Name: torsten.vcf
Type: text/x-vcard
Size: 366 bytes
Desc: not available
Url : http://public.kitware.com/pipermail/insight-users/attachments/20060816/b5e49607/torsten.vcf