[ITK-users] 3D registration problem
Bradley Lowekamp
blowekamp at mail.nih.gov
Wed Jul 16 10:03:38 EDT 2014
Have you looked at the histograms of the two image with the parameters provided to the metric?
Perhaps this example will be of use:
http://www.itk.org/Doxygen/html/RegistrationITKv3_2ImageRegistrationHistogramPlotter_8cxx-example.html
Brad
On Jul 16, 2014, at 9:52 AM, Iyas Hamdan <iyas.hamdan at gmail.com> wrote:
> Hello,
>
> Thanks for your reply!
>
> It's true I haven't shared alot of information about the methode I'm using sorry about that.
>
> My code is based on the example "ImageRegistration20.cxx" with some modifications;
>
> So I'm basically using a 3D Affine registration with Mattes mutual information being the metric and the regular step gradient descent as the optimizer and finally a linear interpolator.
>
> For the initialization, its good and there's no problem here.
>
> But for the result, concerning the registered volume, the scale is good as well but the problem is that the its not aligned with the fixed volume.
>
> And for the parameters, I have only a few ones for the metric ( number of histogram bins and spatial samples ) and I dont think they would have that much of effect on the output, but I tried to change them and I ended up always with almost the same result.
> And for the optimizer, same as before I tried so many values for the relaxation factor and the max/min step lengths as well.
>
> So the problem is not the scale nor the parameters but the registration itself; since it can't find, for each slice of the fixed volume, the slice in the moving volume that corresponds the maximum. So on the output I have a volume that has the same scale as the fixed one but they are not aligned and have no mutual information between them I would say. Even though I tried to change the initialization to prevent the local minimas and I'm starting now with a good initialization but still the same result.
>
>
> And what I found wierd is that when I deleted few slices from the end of the moving volume, the registration worked perfectly and I had good results!
>
>
>
> Hope this was clear and these information was enough, and please let me know if you have any other suggestions or advices.
>
> Thanks in advance,
>
> Iyas
>
>
> On Wed, Jul 16, 2014 at 3:17 PM, Bradley Lowekamp <blowekamp at mail.nih.gov> wrote:
> Hello,
>
> This not the expected behavior for a good registration program.
>
> Is your initialization good?
>
> You haven't shared anything about the registration components you are using. Perhaps you scales are way off? How does the convergence look when monitoring the output?
>
> Perhaps you have too many free parameters? Maybe you are trying to optimize an transform with too many parameters when you are not close. Perhaps you need to do a Translation + Rotation transform such as Euler or Versor first?
>
> Hope that helps,
> Brad
>
> On Jul 15, 2014, at 5:26 AM, Iyas Hamdan <iyas.hamdan at gmail.com> wrote:
>
> > Hello itk users,
> >
> > I'm working on 3D image registration, currently I'm trying to register a CT volume to an MRI volume ( the two volumes have different sizes concerning the number of slices used to construct the 3D volume).
> >
> > I'm doing an affine registration using the mutual information as a metric, and I'm having problems registering those two volumes.
> >
> > What I thiink the problem is, I have a different field of view between both volumes, so when I try to register the whole CT volume to the MRI one the registration falis. However, when I deleted a dozen of slices from the CT volume ( so that I had the same field of view in both volumes) it works just fine and I was able to register the volumes.
> >
> >
> > And by "the registration failed" I meant it worked with no errors but the result was really bad! and that the registered volume was nothing like the fixed one, so if I try to find the difference between each two corresponding slices for example , they were not aligned at all!
> >
> > Is that normal that the registration doesnt work when I have a different field of view in each volume ? or should it deform one volume so that it corresponds to the other one even if they had two different fields of view ?
> >
> > Any help would be really appreciated,
> >
> > Thanks in advance,
> >
> > Iyas
> > _____________________________________
> > Powered by www.kitware.com
> >
> > Visit other Kitware open-source projects at
> > http://www.kitware.com/opensource/opensource.html
> >
> > Kitware offers ITK Training Courses, for more information visit:
> > http://www.kitware.com/products/protraining.php
> >
> > Please keep messages on-topic and check the ITK FAQ at:
> > http://www.itk.org/Wiki/ITK_FAQ
> >
> > Follow this link to subscribe/unsubscribe:
> > http://public.kitware.com/mailman/listinfo/insight-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-users/attachments/20140716/015d3610/attachment-0001.html>
More information about the Insight-users
mailing list