[MINC-users] Re: hires vs. lowres registration

[Andrew Janke]

On 5/6/05, elo@neurorx.com <elo@neurorx.com> wrote:
>
> I'm not sure if mincresample does not have an effect.  For example, the
> process of reducing the resolution and changing coordinates to match the
> T1 image to the T2 space takes place when this tool is run.  So perhaps
> mritoself calculates the correct transformation (.xfm) but could error or
> noise be then introduced afterward during the resampling?

Yes, resampling errors will be introduced when mincresample is run. 
But these same resampling errors are going to occur regardless of
which transformation you apply. In any case, these errors are nothing
to do with registration error.  It is these registration
errors/differences that I believe you are interested in.

> Recently I tested the registration (i.e. mritoself and mincresample) on a
> series of scans with varying quality from different scanners.  In all
> cases registering using the lo-res T1 image worked extremely well for some
> reason.  So if we can be sure that this is happening for a valid reason,
> it would represent a good technique or strategy for us (and perhaps in
> general), during the image registration stage of analysis.

Again, I suspect this is due to the inherrent blurring that is
performed as part of downsampling data.  What you have described here
(low res fitting etc) is one of the fundamental ideas of a hierachial
fitting strategy -- fitting large features first followed by high-res
features. Have a read of Louis Collins' (the author of minctracc)
paper in JCAT regarding this.  (man minctracc will spit up the
reference).

       [2]  Collins  DL,  Neelin  P,  Peters  TM, and Evans AC, "Automatic 3-D
       intersubject  registration  of  MR  volumetric  data  in   standardized
       Talairach space", J. Comput. Assis. Tomogr, 18(2):192-205.

-- 
Andrew Janke      (a.janke@gmail.com || www.cmr.uq.edu.au/~rotor)
Australia->Brisbane     H: +61 7 3390 6332  || M: +61 4 2138 8581