[Insight-users] Parameter Tunning - Help Needed !!!

[Luis Ibanez]

Hi Sateesh,

Well, being pragmatic,
if the two datasets are 'perfectly aligned'
by bother running a registration ?

You could just set up an  itk::IdentityTransform in
an itk::ResampleImageFilter. The filter will take
into account the pixels spacing of both images and
do the rescaling for you.

That will be a more straight forward approach for
solving your problem.



But....
if you want to do it the hard way and spend more
time I would suggest you to do the following:

1) Increase the number of iterations per level to ~2000.

2) Reduce the Max Step Length to ~0.1

3) Increase the Translate scale to ~100000

4) Add Command/Observers to the registration and
    print out the transform parameters at each
    iteration (inside the levels).

5) Plot the components of the rotational matrix
    as they are computed at each iteration.


The fact that you are getting some blank slices
suggests that right now your transform is shrinking
the image. You may have to look closely at the
diagonal elements of the rotational matrix in your
transform. When this elements get values over 1.0
that may indicate a scaling change resulting in
shrinking (since this is the transform from the
fixed image space to the moving image space).




Regards,


    Luis


---------------
cspl wrote:
> Hi Luis,
>  
>    Well as u said, the data sets align perfectly even before 
> registration. That is first slice of CT Volume align perfectly with 
> first slice of PET Volume. Similarly with the last slices of both.
>    Now the problem is with the resolution. CT has 512x512 resolution and 
> PET has 128x128. So we can't fuse these two data sets directly because 
> of difference in resolution. So first i register these two data sets so 
> that i get registered output of dimension 512x512. Then i fuse CT and 
> registered output.
>  
>    Now in registred output i get some blank slices in front and back 
> side of the volume which should not happen because the original data 
> sets are alrealy aligned (if resolution is not considered).
>  
>    Now in this case which paremeter i have to change to remove blank slices.
>    My paremeters are:-
>       Translate scale : 10000
>       Max Step Length : 1
>       Min step Lenght : 0.0001
>       Optimizer scale : 1
>       Spatial samples : 1000
>       bins            : 50
>       Levels       : 3
>       Iterations      : 150
>  
> please help me in this regard.
>  
> Thanks in advance.
>  
> -Regards,
>   Sateesh.