[Insight-users] Mattes MI: "Too many samples map outside..."

[Christoph Palm]

But differences due to endian format is very easy to recognize,
if you look at your images, isn't it? Additionally, the result
of registration might be crap, but the position of the spatial samples 
(without using Padding values and masks) as well as their number should
be independent from the endian format.

-- Christoph



On Tue, 2007-02-13 at 21:50 -0500, Minjie Wu wrote:
> Hi,
> 
> I met the same error when using Mattes MI: "Too many samples map outside...".
> 
> In my case, I resaved both analyze images (moving image and target
> image) using the same the software: ImageJ. Then the error
> disappeared.
> 
> I think the error might due to the endianness of the input images. One
> image is with big endian format, the other one is little endian
> format.
> 
> You may want to try to resave the images using same software, and see
> if this remove the error message.
> 
> Good luck,
> Minjie
> 
> 
> 
> 
> 
> 
> On 2/13/07, Xabier Artaechevarria Artieda <xabiarta at unav.es> wrote:
> > I agree, that is odd. In fact, the images are exactly the same size,
> > and they have been registered with an affine transform previously.
> >
> >
> >
> >
> > Karthik Krishnan <karthik.krishnan at kitware.com> ha escrito:
> >
> > > That's odd.
> > >
> > > Is the moving image a lot smaller than the fixed image ? Samples are taken
> > > from the fixed image. If the moving image is a lot smaller, that could
> > > explain their mapping outside the moving image region.
> > >
> > > -karthik
> > >
> > >
> > > On 2/12/07, Xabier Artaechevarria Artieda <xabiarta at unav.es> wrote:
> > >>
> > >> Hi Karthik,
> > >> Thanks for your answer.
> > >> I just set the maximum number of iterations to 174 and applied the
> > >> resulting transform to the moving image. Comparing it to the fixed
> > >> image, the overlapped seemed good.
> > >> I tried changing the number of spatial samples (from 1/10 to 1/9) and
> > >> now the registration is working fine. I must admit I do not really
> > >> know why.
> > >> Regards,
> > >> Xabi
> > >>
> > >>
> > >>
> > >>
> > >> Karthik Krishnan <karthik.krishnan at kitware.com> ha escrito:
> > >>
> > >>> On 2/9/07, Xabier Artaechevarria Artieda <xabiarta at unav.es> wrote:
> > >>>>
> > >>>> Hallo Christoph,
> > >>>> Thanks for the answer.
> > >>>> I checked after 174 iterations and the overlap seems in fact good,
> > >>>
> > >>>
> > >>> How did you ascertain the fact that the overlap is good ? (I presume you
> > >>> checked the transform, but I don' tsee it printed in the log). There
> > >> must be
> > >>> a significant drift in teh transform from teh initial position to the
> > >>> position after 174 iterations to prompt this error.
> > >>>
> > >>> DId you try resampling with the transform and overlaying the images to
> > >> make
> > >>> sure that the transforms are headed in the right direction ?
> > >>>
> > >>> and
> > >>>> it is considerably better than at the first iteration. Do you think I
> > >>>> could solve the problem by using a higher number of spatial samples?
> > >>>> Now I am using one tenth of the total voxels.
> > >>>
> > >>>
> > >>> 1/10th for MI is plenty.
> > >>>
> > >>>