[MINC-users] minctracc with mask: objective function startsat0.0: what to do?

[Audette, Michel]

Hi Andrew, 

looks like a good plan! 

Thanks again, 

Michel


-----Original Message-----
From: minc-users-bounces at bic.mni.mcgill.ca on behalf of Andrew Janke
Sent: Fri 1/18/2008 4:19 PM
To: MINC users mailing list
Subject: Re: [MINC-users] minctracc with mask: objective function startsat0.0: what to do?
 
> What I am doing, is using a mask to weight an affine transformation to part of an anatomy. For example, I make a mask that coincides with a facial nerve, a 2nd mask that coincides with the ear-drum, and so on. In all of these cases, I want to use a global affine transform, for the whole inner ear, as a starting point for a piecewise affine registration, where one piece is the nerve, another is the ear-drum and so on. I am in fact registering a source of micro-MR, from which the Robert Funnell inner ear model is derived, to high-res patient CT, in order to insert the components of the Funnell model in patient data.
>
> So it isn't mask to mask per se. Also, the mask was in fact dilated with mincmorph (8 successive dilations), in relation to the tight shape that I had. Still getting a null objective function when I try the piecewise.
>
> minctracc -model_mask facialPatchedD8_xyz_rightear_mi_affine_flip9.mnc -mi -w_translations 0.16 0.16 0.16 -w_rotations 0.0028 0.0028 0.0028 -step 0.64 0.64 0.64 -transformation HumanS16885_Stack_nuc_xyz_rightear_mi_affine_9.xfm HumanS16885_Stack_nuc_xyz_res_rigid.mnc.gz etzold_emma_12792115_003002_mri_xyz.mnc HumanS16885_Stack_nuc_xyz_rightear_mi_affine_facialMask9.xfm -clob -lsq6
>
> Does your solution still apply, do you think?

Nope.  (Well not in much of a practical sense in any case... :)

I suspect what is happening here is that the step size across the mask
piece is too large?  ie: how many lattice points are there going to be
with a step size of 0.64 across the masked structure?

Remember that minctracc only evaluates the similarity function for
these points (for speed). Perhaps reduce the step a bit?

The other trick that I have used on occasion with "misbehaving masks"
is that instead of using masking in minctracc I do this:

   1. Make a mask of the thing you want to register.

   2. Pick a reasonable size blurring kernel (ie: if your structure is
5mm across, 1mm sounds good.

   3. Dilate your mask by approx 1/2 the FWHM of step #2.

   4. blur the mask using mincblur with the FWHM from #2.

   5. Multiply the image you want to register with the mask thus
"isolating" the structure

   6. register without masks.

The "soft" edges that this produces tends to be more successful than
the "hard" edge of the mask.

Here's hoping this is of more use to you! :)


-- 
Andrew Janke   (a.janke at gmail.com || http://a.janke.googlepages.com/)
Canberra->Australia    +61 (402) 700 883
_______________________________________________
MINC-users at bic.mni.mcgill.ca
http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users