[MINC-users] Affine transformation matrix

[Alex Zijdenbos]

My guess would be that if the source and the target are in fact
different in scale (such as the infamous Talairach brain), that you
may have trouble fining a good fit this way (but I suspect Aaryani
does have the same scale brains so this may very well work).

Alternatively, I have removed scale from an xfm by concatenating it
with an xfm with the inverse scale, see the snippet of perl code
below.

-- A

Here

&Run() is a glorified system() call
&TmpFile() generates a temporary filename
&CoM() calculates the center of mass of the source volume

---

my @params;
&Run(['xfm2param', $tempxfm], \@params) && die;
chomp @params;

my @scale = (1.0, 1.0, 1.0);
foreach my $param (@params) {
    if ($param =~ /-scale\s+(\S+)\s+(\S+)\s+(\S+)/) {
	 $scale[0] *= $1;
	 $scale[1] *= $2;
	 $scale[2] *= $3;
    }
}

@scale = map { 1.0/$_ } @scale;

my $unscalexfm = &TmpFile('unscale.xfm');
&Run(['param2xfm', '-clobber', '-center', &CoM($CurSource), '-scales',
@scale, $unscalexfm]) && die;




On Tue, Jun 21, 2011 at 10:48 PM, Claude LEPAGE
<claude at bic.mni.mcgill.ca> wrote:
>
> Andrew,
>
> Why not do a -lsq12 transform while disabling scales via the
> -w_scales parameter to minctracc?
>
> Claude
>
>
>>
>> Hi Aaryani,
>>
>> You are asking a pretty loaded question here. If the scaling is small
>> (say 1.001) then I wouldn't expect removing it will make much
>> difference, of course there will still be a difference but it probably
>> will be tolerable.
>>
>> Remember also that an .xfm is really just an affine transformation
>> matrix that can be expressed in a many different combinations of
>> rotations, translations, scales and shears. xfm2param just gives you
>> one of these solutions based upon a set of assumptions (rotations all
>> less than 90deg, etc).
>>
>> As for changing the rotations and translations accordingly, the answer
>> is definitely yes, but the matrix algebra involved in this is going to
>> get pretty ugly pretty quick and in any case will not be the same as
>> the original transformation once the scaling is removed. My suggestion
>> if you really want to do this is to recover the transformation type
>> you want using minctracc from the outset. In your case this would be
>> done via the -lsq6 argument, or perhaps lsq7.
>>
>> a
>>
>>
>>
>> On Wed, Jun 22, 2011 at 10:22, Aaryani Tipirneni <atipirne at stanford.edu> wr=
>> ote:
>> > =A0I have a basic question - As we know, If we do affine registration usi=
>> ng mritotal to the atlas, we will get a .xfm file with the translations, ro=
>> tations, scaling and shearing. If I now want to remove the scaling from the=
>>  .xfm matrix so that the coregistered data has the same scaling as the orig=
>> inal data (to avoid stretching and shrinking), do you expect to still coreg=
>> ister well? Do I have to change even the rotations and translations accordi=
>> ngly?
>> _______________________________________________
>> MINC-users at bic.mni.mcgill.ca
>> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>>
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>