[Insight-users] Re:Registration
CSPL
affable at hd2 . dot . net . in
Wed, 12 Nov 2003 13:21:40 +0530
This is a multi-part message in MIME format.
------=_NextPart_000_000C_01C3A91F.E87221A0
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
Dear Mr.Luis,
ThankYou for your reply. I am enclosing the details about registration =
components.
Transform : CenteredAffineTransform
Optimizer : RegularStepGradientDescentOptimizer
Metric :MattesMutualInformationImageToImageMetric
Interpolation :LinearInterpolateImageFunction
Registration Method : MultiResolutionImageRegistrationMethod
Pyramid : RecursiveMultiResolutionPyramidImageFilter
Regards,
SriValli.
Your reply:
Hi Srivalli,
Here are some of the points to check
0) Make sure that the pixel spacing
and origin of your image are correctly
set in millimeters.
1) What registration components are you
using ?
- optimizer ?
- metric ?
- transform ?
2) Connect an Observer to the optimizer
and print out the metric value and
the transform parameters as the registration
progresses.
3) Plot the Metric values versus the number
of iterations and evaluate how noisy the
metric is, and how much it is increasing.
The plots of Metric and transform parameters
versus number of iterations are the best
guidance for determining how to tune the
parameters of a registration method.
Please provide more details about the components
you are combining.
Thanks
Luis
My question:
Dear Mr.Luis,
I am working on medical image registration. I could get the single =
modality image registration well.=20
But,Facing problems in MultiModalityRegistration.
In my application I have to register MR and SPECT volumes, MR as FIXED =
and SPECT as MOVING.
Both MR and SPECT are PreProcessed.=20
MR is of size 256X256X120.
Brain is extracted from skull in MR volume and the extracted brain is =
used as FIXED.
ITK Neighbourhoodconnected filter is used to extract object based on the =
given seed point and threshold range.=20
SPECT is processed to make cubic voxels.
Procesed to find the minimum dimension of SPECT and creates a new =
volume with all dimensions equal to minimum dimension.
Inititally SPECT is of size 128X128X49.After processing it will become =
128X128X96.Voxel Width, Voxel Height and Voxel Depth will be same.
After preprocessing extracted brain and spect will be given as inputs =
for registration.
To view the output of registration, I am merging SPECT with original MR =
volume(includes both brain and skull) so that spect exactly fits in the =
brain part.
But, when I see the output spect is not positioning correctly.
Sometimes I see that middle slices are ok but,failing at last slices.
I tried in many ways by chaning the parameters.=20
I feel that registration depends on fixed and moving images. Parameter =
tuning can be succes only if input images are correct.
Can you please tell us what could be the reason to fail?
I am doubt whether I am wrong in giving input images or in tuning =
parameters.
------=_NextPart_000_000C_01C3A91F.E87221A0
Content-Type: text/html;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META http-equiv=3DContent-Type content=3D"text/html; =
charset=3Diso-8859-1">
<META content=3D"MSHTML 6.00.2600.0" name=3DGENERATOR>
<STYLE></STYLE>
</HEAD>
<BODY bgColor=3D#ffffff>
<DIV><FONT face=3DArial size=3D2>Dear Mr.Luis,</FONT></DIV>
<DIV> </DIV>
<DIV><FONT face=3DArial size=3D2> ThankYou for your reply. I am =
enclosing the=20
details about registration components.</FONT></DIV>
<DIV> </DIV>
<DIV><FONT face=3DArial size=3D2>Transform : =
CenteredAffineTransform<BR>Optimizer :=20
RegularStepGradientDescentOptimizer<BR>Metric=20
:MattesMutualInformationImageToImageMetric<BR>Interpolation=20
:LinearInterpolateImageFunction<BR>Registration Method :=20
MultiResolutionImageRegistrationMethod<BR>Pyramid :=20
RecursiveMultiResolutionPyramidImageFilter</FONT></DIV>
<DIV> </DIV><FONT face=3DArial size=3D2>
<DIV><BR>Regards,<BR>SriValli.<BR></DIV>
<DIV>Your reply:</DIV>
<DIV>Hi Srivalli,</DIV>
<DIV> </DIV>
<DIV>Here are some of the points to check</DIV>
<DIV> </DIV>
<DIV><BR>0) Make sure that the pixel spacing<BR> and =
origin of=20
your image are correctly<BR> set in millimeters.</DIV>
<DIV> </DIV>
<DIV><BR>1) What registration components are you<BR> =20
using ?</DIV>
<DIV> </DIV>
<DIV> - optimizer ?<BR> -=20
metric ?<BR> - transform ?</DIV>
<DIV> </DIV>
<DIV><BR>2) Connect an Observer to the optimizer<BR> =
and print=20
out the metric value and<BR> the transform parameters =
as the=20
registration<BR> progresses.</DIV>
<DIV> </DIV>
<DIV>3) Plot the Metric values versus the number<BR> =
of=20
iterations and evaluate how noisy the<BR> metric is, =
and how=20
much it is increasing.</DIV>
<DIV> </DIV>
<DIV><BR>The plots of Metric and transform parameters<BR>versus number =
of=20
iterations are the best<BR>guidance for determining how to tune=20
the<BR>parameters of a registration method.</DIV>
<DIV> </DIV>
<DIV>Please provide more details about the components<BR>you are=20
combining.</DIV>
<DIV> </DIV>
<DIV><BR>Thanks</DIV>
<DIV> </DIV>
<DIV><BR> Luis</DIV>
<DIV> </DIV>
<DIV>My question:</DIV>
<DIV>Dear Mr.Luis,</DIV>
<DIV> </DIV>
<DIV> I am working on medical image registration. I could get the =
single=20
modality image registration well. <BR>But,Facing problems in=20
MultiModalityRegistration.<BR>In my application I have to register MR =
and SPECT=20
volumes, MR as FIXED and SPECT as MOVING.<BR>Both MR and SPECT are =
PreProcessed.=20
<BR>MR is of size 256X256X120.<BR>Brain is extracted from skull in MR =
volume and=20
the extracted brain is used as FIXED.<BR>ITK Neighbourhoodconnected =
filter is=20
used to extract object based on the given seed point and threshold =
range. </DIV>
<DIV> </DIV>
<DIV>SPECT is processed to make cubic voxels.<BR>Procesed to find the =
minimum=20
dimension of SPECT and creates a new volume with all =
dimensions=20
equal to minimum dimension.<BR> Inititally SPECT is of size=20
128X128X49.After processing it will become 128X128X96.Voxel Width, Voxel =
Height=20
and Voxel Depth will be same.</DIV>
<DIV> </DIV>
<DIV>After preprocessing extracted brain and spect will be given as =
inputs for=20
registration.</DIV>
<DIV> </DIV>
<DIV>To view the output of registration, I am merging SPECT with =
original MR=20
volume(includes both brain and skull) so that spect exactly fits in the =
brain=20
part.<BR>But, when I see the output spect is not positioning=20
correctly.<BR>Sometimes I see that middle slices are ok =
but,failing at=20
last slices.<BR>I tried in many ways by chaning the parameters. <BR>I =
feel that=20
registration depends on fixed and moving images. Parameter tuning can be =
succes=20
only if input images are correct.<BR>Can you please tell us what could =
be the=20
reason to fail?<BR>I am doubt whether I am wrong in giving input =
images or=20
in tuning parameters.<BR></DIV></FONT></BODY></HTML>
------=_NextPart_000_000C_01C3A91F.E87221A0--