<br>Hi Ganesh,<br><br>Thanks for describing the clinical constraints of your application.<br><br>Unfortunately the pair of 2D images do not have enough information <br>to be considered equivalent to a 3D image. Should that be the case,<br>
we would not need CT scanners, and would have been fine with<br>dual plane X-Ray machines.<br><br>but, ... for the sake of being pragmatic, you can certainly claim that<br>the 2D images carry "some" valuable information that allows you to<br>
"estimate" a "reasonable" deformation in 3D space for a small price.<br><br>For example, if you have patient bending over her/his abdomen <br>between the T1 and T2 images, then you could get most of this<br>
"bending" information from the YZ plane (using LPS coordinates):<br><a href="http://www.cmake.org/Wiki/Proposals:Orientation#DICOM_LPS_Orientation_-_Full_Body_Graphical_Example">http://www.cmake.org/Wiki/Proposals:Orientation#DICOM_LPS_Orientation_-_Full_Body_Graphical_Example</a><br>
<br>You could project that 2D deformation field in 3D, (by simply adding<br>that the displacement along X is zero), and apply such field to the<br>3D image using the itkWarpImageFilter.<br><br>Note that in that "restricted bending" case, the image in plane XZ<br>
will not be of much use, since tissue in the patient's back will move<br>upwards, while tissue in the patient's front will move downwards<br>in that plane, and the projection image will hardly allow you to<br>disambiguate these two opposite displacements.<br>
<br>So, you probably can use the YZ plane for a "coarse" estimation<br>of deformations such as Breathing, and bending.<br><br><br><br>You should be very careful not to extrapolating those estimations<br>for critical application such as image guided therapy....<br>
<br><br>The case of a "general deformation" certainly can't be solve with<br>
just a pair of 2D projection images.<br>
<br><br> Regards,<br><br><br> Luis<br><br><br>------------------------------------------------------------------------------------------<br><div class="gmail_quote">On Thu, Apr 8, 2010 at 3:58 PM, Ganesh Narayanasamy <span dir="ltr"><<a href="mailto:nganesh76@hotmail.com">nganesh76@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<div>
<br>The main reason for not doing a 3D-3D registration is that we do not, obviously, have the luxury of acquiring 3D data sets every time the patients come by. The 3D-CT scan is acquired once with which radiotherapy planning is done initially. After that visit, the patients are scanned daily using a kV-imager twice that gives two 2D images in orthogonal planes (as described below).<br>
I think these 2 orthogonal projection images have enough information to be considered "equivalent" to a 3D image. Am i right ? If not, then why not?<br><br>When these 2 orthogonal projection images acquired on 2 different days are registered to each other, I would like to obtain the corresponding deformation fields in XZ and YZ. For every point P in the fixed image, I want to evaluate its new location P' in the moving image (using say, WarpImageFilter).<br>
<br><br>Thank you,<div class="im"><br>Ganesh<br><br><font color="#1f497d" face="Verdana">************************************************************************</font><font color="#1f497d" face="Verdana"><br></font><font color="#1f497d" face="Verdana">
"What the mind of man can conceive and believe, It can achieve" - </font><font color="#1f497d" face="Verdana"><span>Napolean</span></font><font color="#1f497d" face="Verdana"> </font><font color="#1f497d" face="Verdana"><span>Hill</span></font><font color="#1f497d" face="Verdana"><br>
</font><font color="#1f497d" face="Verdana">
-Ganesh
Narayanasamy,Ph.D., PostDoc-University of Kentucky Cancer Center</font><font color="#1f497d" face="Verdana"><br></font><font color="#1f497d" face="Verdana">
Earlier work:
<a href="http://www.sciencedaily.com/releases/2007/05/070504133017.htm" target="_blank">http://www.sciencedaily.com/releases/2007/05/070504133017.htm</a></font><font color="#1f497d" face="Verdana"><br></font><font color="#1f497d" face="Verdana">
************************************************************************</font><br><br><br><br></div>> Date: Thu, 8 Apr 2010 14:43:28 -0400<br>> Subject: Re: [Insight-users] Applying two 2D registration transforms into a 3D image<br>
> From: <a href="mailto:luis.ibanez@kitware.com" target="_blank">luis.ibanez@kitware.com</a><br>> To: <a href="mailto:nganesh76@hotmail.com" target="_blank">nganesh76@hotmail.com</a><br>> CC: <a href="mailto:insight-users@itk.org" target="_blank">insight-users@itk.org</a><div>
<div></div><div class="h5"><br>> <br>> Hi Ganesh,<br>> <br>> Thanks for the detailed description of the problem that you are working on.<br>> <br>> Unfortunately, the deformation fields of the two 2D deformable registrations<br>
> do not carry enough information for you to do a 3D deformation of the T1<br>> and T2 original 3D-images.<br>> <br>> <br>> Any reason for not doing real 3D deformable registration in the<br>> original images ?<br>
> <br>> [...ITK will do registration on N-D images...]<br>> <br>> <br>> Please let us know,<br>> <br>> <br>> Thanks<br>> <br>> <br>> Luis<br>> <br>> <br>> <br>
> ------------------------------------------------------------------------------------<br>> On Wed, Apr 7, 2010 at 1:06 PM, Ganesh Narayanasamy<br>> <<a href="mailto:nganesh76@hotmail.com" target="_blank">nganesh76@hotmail.com</a>> wrote:<br>
> ><br>> > I have a pair of 3D-CT images acquired on a phantom with some internal<br>> > changes (known).<br>> > A 2D-DRR (digitally reconstructed radiograph) image is constructed such<br>> > that the pixel value on the DRR image should be given by the accumulated<br>
> > intensities of the 3D image volume along the direction. With the T1 or fixed<br>> > 3D-CT image, we have created two such DRRs, say along the XZ and YZ axes.<br>> > Let us call these 2 DRRs as T1DRR_XZ and T1DRR_YZ.<br>
> ><br>> > Likewise, we have created 2 DRRs from the T2 3D-CT image namely, T2DRR_XZ<br>> > and T2DRR_YZ.<br>> ><br>> > Using a deformable 2D registration model (say, Bspline), we would like to<br>
> > register T1DRR_XZ with T2DRR_XZ and also T1DRR_YZ with T2DRR_YZ.