[Insight-users] problems on deformable registration 8
Sean Lee
kevinseanlee at live.com
Thu Jul 9 16:44:10 EDT 2009
Hi, Luis,
I found that another parameter could affect the result. If I change gridSizeOnImage.Fill( 8 ) to gridSizeOnImage.Fill( 10 ), it could result in the case with 59 slices that used to run 34 iterations only run one iteration.
So it is really depending on the parameters you input.
Thanks.
Sean
From: kevinseanlee at live.com
To: luis.ibanez at kitware.com
Date: Thu, 9 Jul 2009 12:53:39 -0500
CC: insight-users at itk.org
Subject: Re: [Insight-users] problems on deformable registration 8
Hi, Luis,
I am trying to provide more information and tested it with several other numbers. Here is the results.
29, 56, 57, 58, 60 slices: 1 iteration with a bad result.
30 slices; 45 iterations with a good result. (of course, only half volume is matched)59 slices; 34 iterations with a good result.
Let me know what I can do for this issue.
Thanks.
Sean
From: kevinseanlee at live.com
To: luis.ibanez at kitware.com
Date: Thu, 9 Jul 2009 08:22:59 -0500
CC: insight-users at itk.org
Subject: Re: [Insight-users] problems on deformable registration 8
Hi, Luis,
Thanks for your reply.
> A) Did you make *any* changes to the source code
> of the DeformableRegistration8.cxx example ?
I believe I did not.
> B) How are you reading your images ?
> Are they DICOM files ?
> Are you using a Series Reader ?They are raw data with mhd or mha header files and I used following filter to read them. VolumeReaderType::Pointer fixVolumeReader = VolumeReaderType::New(); fixVolumeReader->SetFileName(fixVolumeFileName); fixVolumeReader->Update();
> C) How *exactly* have you been removing slices from them.
My original volume are 512X512. First I down-sampled them to 256X256 with itk::ResampleImageFilter and thenrun itk::ExtractImageFilter with following setting Input3DImageType::RegionType sliceRegion; Input3DImageType::IndexType sliceStartPoint; sliceStartPoint[0] = 0; sliceStartPoint[1] = 0; sliceStartPoint[2] = 0; Input3DImageType::SizeType NewVolumeSize; NewVolumeSize[0]=fixVolumeSize[0]; NewVolumeSize[1]=fixVolumeSize[1]; NewVolumeSize[2]=59; //or 60 !!!!! here is the way I selecting the number of slices !!!!! sliceRegion.SetSize(NewVolumeSize); sliceRegion.SetIndex(sliceStartPoint);It is my understanding that the volume size should not affect the results that much. But like what I experienced earlier, changing that number could result in reaching "Convergence: function tolerance reached." without a good result. I hope it can always reach "Convergence: gradient tolerance reached."
Thanks.
Sean
> Date: Wed, 8 Jul 2009 18:08:48 -0400
> From: luis.ibanez at kitware.com
> To: kevinseanlee at live.com
> CC: insight-users at itk.org
> Subject: Re: [Insight-users] problems on deformable registration 8
>
>
> Hi Sean,
>
>
> The Exception
>
> itk::InvalidRequestedRegionError
>
>
> doesn't seem to be related to the registration process.
>
>
> A) Did you make *any* changes to the source code
> of the DeformableRegistration8.cxx example ?
>
>
> B) How are you reading your images ?
> Are they DICOM files ?
> Are you using a Series Reader ?
>
>
> C) How *exactly* have you been removing slices from them.
>
>
> D) 2Gb of RAM should be enough for registering two images of
> 64 slices with 256x256 pixels each.
>
>
>
>
> Something else is going on here....
>
>
>
> Please let us know about (A),(B) and (C)
>
>
>
> Thanks
>
>
>
> Luis
>
>
>
>
>
> ------------------
> Sean lee wrote:
> > Hi, Luis,
> >
> > Thank you for your reply.
> >
> > I have 2GB RAM. Here is the error message if I run more than 62 slices:
> > Unhandled exception at 0x7c812afb in RegTest.exe: Microsoft C++
> > exception: itk::InvalidRequestedRegionError at memory location 0x0013fbbc..
> >
> > My CT volumes are liver images from two different phases.
> >
> > I also did another test on these two volumes by switch them (switch
> > fixed and moving volumes). If I run 62 slices,, it would come with same
> > memory problem (see above error message). This time if I run 60 slices,
> > it would reach "CONVERGENCE: REL_REDUCTION_OF_F <= FACTR*EPSMCH" after
> > one iteration; but it seems 59 slices allow me to run 34 iterations and
> > gave me a good result.
> >
> > So, it seems the good results came with "reach Projected Gradient
> > Tolerance". If it reaches "CONVERGENCE: REL_REDUCTION_OF_F <=
> > FACTR*EPSMCH" first, it come with a "bad" result. How can I control this
> > situation?
> >
> > Thanks.
> >
> > Sean
> >
> > > Date: Wed, 8 Jul 2009 17:39:25 -0400
> > > From: luis.ibanez at kitware.com
> > > To: kevinseanlee at live.com
> > > CC: insight-users at itk.org
> > > Subject: Re: [Insight-users] problems on deformable registration 8
> > >
> > >
> > > Hi Sean,
> > >
> > > This is very suspicious.
> > >
> > >
> > > A) your images do not seem to be that big...
> > >
> > > One image of:
> > >
> > > 64 slices of 256 x 256 in 16bits
> > >
> > > will take 8 Mb
> > >
> > >
> > > How much RAM do you have ?
> > >
> > >
> > > B) Just removing one slice (out of 60) from your
> > > dataset shouldn't lead the registration to
> > > behave differently.
> > >
> > >
> > > Unless...
> > >
> > > There is some special content in that slice...
> > >
> > > Are any objects in Slice 60 that have very
> > > different intensities to those objects visible
> > > in slice 59 ?
> > >
> > >
> > > ...and...
> > >
> > >
> > > C) How are you removing that slice ?
> > >
> > > Are you using a region of interest extraction first ?
> > >
> > >
> > >
> > >
> > > Please let us know,
> > >
> > >
> > > Thanks
> > >
> > >
> > >
> > > Luis
> > >
> > >
> > >
> > > ---------------
> > > Sean lee wrote:
> > > > Hello,
> > > >
> > > > I am trying to use the example of deformable registration 8 on two ct
> > > > image volumes (64 256X256 slices). The problem is If i use more
> > than 62
> > > > slices, I would have memory problem; I run 60 slice, I got a perfect
> > > > results with 20 iterations; but if I run 59 slices, it only run one
> > > > iteration and reaches "CONVERGENCE: REL_REDUCTION_OF_F <=
> > FACTR*EPSMCH".
> > > > I tried to reduce the Cost Function Convergence Factor to 1e+1 and it
> > > > still comes same results. What could be the problem and any suggested
> > > > solutions? Thanks. Sean
> > > > >
> > > >
> > > >
> > ------------------------------------------------------------------------
> > > >
> > > > _____________________________________
> > > > Powered by www.kitware.com
> > > >
> > > > Visit other Kitware open-source projects at
> > > > http://www.kitware.com/opensource/opensource.html
> > > >
> > > > 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://www.itk.org/mailman/listinfo/insight-users
> >
> > ------------------------------------------------------------------------
> > Windows Live™ SkyDrive™: Get 25 GB of free online storage. Get it on
> > your BlackBerry or iPhone.
> > <http://windowslive.com/online/skydrive?ocid=TXT_TAGLM_WL_SD_25GB_062009>
Hotmail® has ever-growing storage! Don’t worry about storage limits. Check it out.
Windows Live™ SkyDrive™: Get 25 GB of free online storage. Get it on your BlackBerry or iPhone.
_________________________________________________________________
Insert movie times and more without leaving Hotmail®.
http://windowslive.com/Tutorial/Hotmail/QuickAdd?ocid=TXT_TAGLM_WL_HM_Tutorial_QuickAdd_062009
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090709/01421bfb/attachment-0001.htm>
More information about the Insight-users
mailing list