[Insight-users] Problem on PET-CT Registration - Help Needed!!!
Luis Ibanez
luis.ibanez@kitware.com
Fri, 18 Apr 2003 09:41:08 -0400
Hi Sateesh,
If the registration is not performing a single
iteration then you certainly have an initialization
problem.
Things to check for:
1) Are you setting the Initial transform parameter ?
Please print out the transform parameters as they
are just before starting the registration...
That will throw some light on the problem.
2) If you use the CenteredAffineTransform, it is
important to also use the
CenteredTransformInitializer
Centered transforms are not as intuitive for
initialization.
Examples are available on the SoftwareGuide
http://www.itk.org/ItkSoftwareGuide.pdf
take a look at Section 7.5.3 starting in
pdf-page 198
The associated examples are in
Insight/Examples/Registration
Regards,
Luis
-------------------
cspl wrote:
> Hi Lius,
> Thanks for ur reply.
> As u said i added an Observer and tested the registration code. This time
> i started with a 2d images. My inputs are one CT (512x512x1) as fixed and
> one PET (128x128x1) to the registrations application. Both pixel type is of
> short. Pixel spacing of CT is 0.976562\0.976562 and of PET image is
> 3.906250\3.906250.
>
> When i run the application, it is not performing atleast one iteration
> and it generates a itk error same as before.
>
> itk::ExceptionObject (0012FC2C)
> Location: "Unknown"
> File:
> G:\NEWITKFILES\NewVersionITK7_4_2003\itksrc\Code\Algorithms\itkMattesMutualI
> nformati
> onImageToImageMetric.txx
> Line: 546
> Description: itk::ERROR:
> MattesMutualInformationImageToImageMetric(00302050): Too many sam
> ples map outside moving image buffer: 0 / 1000
>
> I used the following settings:-
> No. of iterations: 150
> Translate Scale : 10000
> Optimizer Scale : 1
> Max Step Length : 4
> Min Step Length : 0.01
> Spatial Samples : 1000
> Bins : 50
> No. of levels : 5
>
> I am using "CenteredAffineTransform" for transformation. For this i am
> shifting the origin to the image center for both the images.
>
> I added shrinking factor of 4 for both the images but no change in the
> output.
>
> What could be the problem??? Is this because of the great change in the
> image dimensions?? But when we find the image dimensions in terms of
> millimeters we get almost same dimensions of both of images (image
> dimensions x pixel spacing). In this case both are of same size. Then what
> could be the problem.
>
> I could not able to find a proper values for registration parameters.
>
> Please suggest some settings for it.
>
> Thanks in advance.
>
> -Regards,
> Sateesh.
>
> ----- Original Message -----
> From: "Luis Ibanez" <luis.ibanez@kitware.com>
> To: "cspl" <affable@hd2.dot.net.in>
> Cc: <insight-users@public.kitware.com>
> Sent: Thursday, April 17, 2003 08:45 PM
> Subject: Re: [Insight-users] Problem on PET-CT Registration - Help Needed!!!
>
>
>
>>Hi Cspl,
>>
>>1) Did you plug an Observer to your registration ?
>>
>> Are you printing out values of the transform
>> parameters as the registration progesses ?
>>
>> Tracking the evolution of the transform will
>> give you the indications you need for tunning
>> parameters.
>>
>>2) Whenever you get the exception regarding the
>> low number of samples, it simply means that
>> the overlap between the two images is too small.
>>
>> In other words, that the transform parameters
>> diverged and the current transform is mapping the
>> image too far from the expected position.
>>
>> The action to take in this case is to track how
>> long it took for the transfrom to diverge. For
>> this you need to solve (1) [add an observer and
>> print out the transform parameters].
>>
>>
>>3) There is no point in using multiresolution if you
>> dont set the shrinking parameters. You could in
>> that case simply use a normal registration with
>> more iterations.
>>
>>
>>Please add an observer to the registration and track
>>the evolution of the transform. You may want to take
>>a look at the examples in
>>
>> Insight/Examples/Registration/
>>
>>
>>Note that there are also Tcl versions for some of them.
>>
>>
>> Regards,
>>
>>
>> Luis
>>
>>
>>----------------------
>>cspl wrote:
>>
>>>Hi Luis and all,
>>> Well back with the registration problem again.
>>> Let me explain my problem.
>>> I am using PET and CT images as input to registration. PET as fixed
>>
> and
>
>>>CT as moving images. Total number of images are 35 in each set. So my
>>
> data
>
>>>sets are 128x128x35 for PET and 512x512x35 for CT.
>>>Pixel spacing for CT images are 0.976562\0.976562 and for PET images are
>>>3.906250\3.906250.
>>>Now the parameters for registration are
>>> No. of iterations: 150
>>> Translate Scale : 10000
>>> Optimizer Scale : 1
>>> Max Step Length : 4
>>> Min Step Length : 0.01
>>> Spatial Samples : 1000
>>> Bins : 50
>>> No. of levels : 5
>>> I am not using any shrinking factors in the registration.
>>
> SetMinimize()
>
>>>method is used in registration process.
>>>
>>> The above setting are working for other data sets like mri-mri
>>>registration.
>>> These parameters are working for single slice of above data set i.e.
>>
> if i
>
>>>give 1 PET as fixed and 1 CT as moving image with the above parameters
>>
> and
>
>>>perform registration it is working good with out any itk error.
>>>
>>> But if i give 35 slices it is giving the following error message:-
>>> itk::ERROR:MattesMutualInformationImageToImageMetic(01A93ED0):
>>
> Two
>
>>>many samples map outside moving image buffer: 0/1000.
>>>
>>> I have tried with different values for the above parameters but still
>>
> this
>
>>>error is generated.
>>>
>>> If u have any idea of parameters for these data sets please suggest
>>
> me.
>
>>> Please help me in this regard.
>>>
>>> Thanks in advance.
>>>
>>>-Regards,
>>> Sateesh.
>>>
>>>
>>>_______________________________________________
>>>Insight-users mailing list
>>>Insight-users@public.kitware.com
>>>http://public.kitware.com/mailman/listinfo/insight-users
>>>
>>
>>
>>
>>
>
> _______________________________________________
> Insight-users mailing list
> Insight-users@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-users
>