[Insight-users] Re: number of spatial samples

Luis Ibanez luis.ibanez at kitware.com
Tue Jun 8 11:53:17 EDT 2004


Hi Christos,

You probably will enjoy reading the Mattes et al. paper in TMI

"PET-CT Image Registration in the Chest Using Free-form Deformations"
D. Mattes, D. R. Haynor, H. Vesselle, T. Lewellen and W. Eubank
IEEE Transactions in Medical Imaging.
Vol.22, No.1, January 2003. pp.120-128.

-----

Quoting from the Multi-resolution
approach section on page 124:


"We follow the generas rule that the
  number of data point Nm at resolution
  level 'm' should be proportional to the
  number of independent parameters for which
  we are trying to solve

         Nm = Eta * Ndc * Nvv

   Eta = a proportionality constant
   Ndc = is the number of deformation
         coefficients
   Nvv = Number of voxels in the volume

   Finally, the optimizer's termination
   criterion is varied. The iteration terminates
   when the change in the objective function
   between consecutive iterations falls below a
   specified value epsilon. At coarse resolutions,
   this criterion can be relaxed, it is set more
   strictly with increasing resolution.

   The alignment process is divided into two
   registrations: a rigid body   (Ndc = 6 parameters,
   Eta = 0.05) followed by a deformation (Ndc paramters,
   Eta = 1e-04) recovery. "



As you will see from table I in the same page,
the percent of pixels used as sample points range
from 0.8% at the coarsest resolution to 19.7% at the
finest resolution.




Regards,


    Luis



-----------------------------
Christos Panagiotou wrote:

> Dear Luis
> 
> in Mattes implementation of MI the number of spatial samples which the user
> needs to set for the metric is around 20% of the pixels of the 
> (biggest?) image to be registered
> (i assume that in 3D it would be the number of voxels)
> 
> if we use a multi resolution method the number of spatial samples whould 
> be adjusted for each level of the pyramid?
> is this done automaticaly in the multiResolutionRegistrationMethod? do 
> we have to implement it explicitly or do we just
> set 20% of the finest (or lowest?) resolution level (1 1 1)
> 
> 







More information about the Insight-users mailing list