[MINC-users] nonlinear registration in minc

[Jason Lerch]

Ernest Lo wrote:

>Hello,
>
>I would like to perform a nonlinear registration of one image to another.  
>In fact it is the same patient, but there is some minor distortion in one
>of the scans (that is affecting our tissue segmentation results).  So I'm 
>looking for a registration technique that will not only align the images 
>but correct for the distortion.
>
>I have experimented with minctracc -nonlinear and mritoself with no 
>success.  I don't have much experience in this area but it seems to me 
>that a good nonlinear reg algorithm should be able to easily align these 
>two images.  Would anyone know of a good way to do this, in terms of 
>parameter settings, scripts, commands, etc.?  
>
>I have access to a couple of tools as well - nlfit_xcorr.pl and
>nlfit_smr.pl - but these require various model files in addition to the 
>target image (blur levels etc.), which I don't know how to generate.  Any 
>advice in this area would be appreciated as well.
>  
>
There are, as far as I know, no convenient scripts available for subject 
to subject non-linear registration. So you'll have to use minctracc 
itself. The basic options you want are:

-nonlinear: erm, I'm sure you can figure out what this option does.
-xcorr: objective function
-step: 3 values for the spacing of the deformation grid.
-simplex: size of search area.
-lattice_diameter: width of sub-lattice (three values)
-iterations: number of iterations.

There are more that might be useful (weight, stiffness, etc.), but those 
should get you started. Also note that a hierarchical process often 
gives the best results: i.e. align two blurred versions first, then use 
the -transform option to minctracc to initialize a second registration 
at less blurring, and so on.

Good luck,

Jason

>Thanks!
>
>
>Ernest Lo
>NeuroRX Research
>Montreal, Canada
>
>
>
>
>_______________________________________________
>MINC-users at bic.mni.mcgill.ca
>http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>  
>