[MINC-users] spatially-variant mincblur?

[Olivier Rousset]

Salut Louis,

Well I have to say that we are thinking along the same
lines here... great! I will try that -got carried away
by other things lately- and it sounds that this should
work. BTW, are you going to HBM'06 in Italy? I forgot
to let you know that our abstract "PET quantification
using an MRI-based Brain Atlas: Effect of ROI 
definition on Dopamine Binding Potential with and
without Partial Volume correction"  Olivier G.
Rousset1, D. Louis Collins2, Hiroto Kuwabara1, Andrew
H. Crabb1, Dean F. Wong1" was accepted for poster
presentation. Hope to see you there!!

Will keep you posted on the progress on the
non-stationnary PSF. 

Bien à toi, et bonne fin de semaine (bon week-end
quoi).

- Olivier

--- "D. Louis Collins" <louis.collins at mcgill.ca>
wrote:

> Salut Olivier,
> 
> I think that it would be possible to create (or at
> least fake) 
> non-stationary blurring for simulation of pet data.
> 
> I would use masks that define the regions that
> correspond to a 
> particular spatial resolution.  Suppose that you
> define 3 concentric 
> regions, with res=5mm at the center, then 6mm and
> 7mm for the two 
> concentric rings. (note that the ROIs don't have to
> be spherical, but 
> should actually correspond to the measured
> resolution field).
> 
> I would then blur the simulated PET data with 3
> kernels (fwhm=5,6 and 
> 7mm) to yield three blurred volumes.  Now mask these
> volumes with the 
> ROIs defined above and then add the results
> together.  Voila, the 
> resulting volume will have spatially varying
> resolution.
> 
> But wait you say, there will be harsh
> discontinuities at the borders of 
> these ROIs.  Right you are.  So, blur the masks with
> a small (say 
> fwhm=3mm) spatial blurring kernel.  Normalize them,
> so that when added 
> together, all voxels sum to 1.0.
> 
> Now, multiple the 3 blurred PET volumes with the 3
> masks and add them 
> together and you'll have cheap, continuous,
> spatially-variant 
> resolution for your simulated PET volume.
> 
> While this is a hack, I think that it would go a
> long way for system 
> evaluation.  You can probably define 5-7 concentric
> regions and you'd 
> get even better results.  (and recoding mincblur for
> spatially varying 
> kernel is a no-go since it uses Fourier Transforms
> to achieve blurring. 
>   You'd have to completely rewrite it to do
> convolution in the spatial 
> domain - which is not unlike what I describe above!)
> 
> Good luck!
> 
> -Louis
> 
> 
> On Apr 22, 2006, at 5:23 PM, Olivier Rousset wrote:
> 
> > Steve,
> > What I am in search of is a non-stationary version
> of
> > mincblur, which currently supports anisotropic
> > blurring kernels, but provides no spatial
> variation
> > encoding. I would like to create simulated PET
> data,
> > for which I need a spatially-variant point-spread
> > function (PSF) that would be more realistic than a
> > stationary PSF.
> >
> > - Olivier
> >
> > --- "Steve M. Robbins"
> <steven.robbins at videotron.ca>
> > wrote:
> >
> >> On Mon, Apr 17, 2006 at 02:58:50PM -0700, Olivier
> >> Rousset wrote:
> >>> I was wondering whether a non-stationary kernel
> >> can be
> >>> specified with mincblur (apparently not in the
> >> current
> >>> version), or if alternatively there would be a
> >> program
> >>> that could do that? I am thinking that maybe
> >>> stationary kernels could be used on parts of the
> >>> images [where spatial resolution is assumed
> >> constant]
> >>> and then combine the resulting blurred images
> >> [since
> >>> these operations can be considered as linear];
> >> this
> >>> would involve masking of the image with
> >> appropriate
> >>> masks defined using "resolution isocontours".
> >>
> >> It sounds like you are after a non-stationary
> >> isotropic smoothing --
> >> as opposed to a non-isotropic ("don't blur across
> >> edges") smoothing.
> >>
> >> Just out of curiousity: what sort of data
> requires
> >> that treatment?
> >>
> >> -Steve
> >>
> >> _______________________________________________
> >> 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
> >
> 
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
>
http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>