[Insight-users] B-Splines registration with fixed points

Tue Jul 26 07:54:22 EDT 2011

Hi Albert,

It looks like we are not checking for division by zero or something like that, so probably a bug.
(The modified pointset does not have a distance of zero at the start already, and I think your fixed and moving image have identical spacing etc)

Can you also give me your fixed and moving image, so I can reproduce this?

Cheers,

Marius

From: krekbert at gmail.com [mailto:krekbert at gmail.com] On Behalf Of Albert Gubern Mérida
Sent: dinsdag 26 juli 2011 11:38
To: Staring, M. (LKEB)
Cc: insight-users at itk.org
Subject: Re: [Insight-users] B-Splines registration with fixed points

Hi Marius,

I was using version 4.3. I have just upgraded to 4.4 i have already tried what you told me. Here an overview what I do:

1. I obtain the sternum points from the fixed and moving image.
2. I apply a translation in y direction to the moving image. Then both pectoral/sternums are aligned. I do that "manually" not using elastix.
3. I apply the registration code using similar elastix parameter file as the one you sent to me to penalize deviation away from the sternum points. Because the moving input images is already translated, the sternum points is the same as the one in the fixed image. So, the text file which contains the list of points are the same for the fixed and moving images. However, doing that I get the following output ( I attach the log file called elastix_same_points.log. Note in it that moving_points.txt and fixed_points.txt contains exactly the same points):
##########################################################################################################################################

  Sampling gradients ...
  Progress: 100%  Sampling the gradients took 1 Minutes, 24 Seconds
Automatic parameter estimation took 1 Minutes, 25 Seconds
1:ItNr  2:Metric           3a:Time           3b:StepSize     4:||Gradient||    Metric0            Metric1            Time[ms]            ||Gradient0||     ||Gradient1||
0          -0.938266        0.000000         1097.045756   -nan     -0.938266        0.000000         92520  0.001563            -nan
1          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
2          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
3          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
4          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
5          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
6          -nan     0.000000         1097.045756   -nan     0.000000         -nan     720      0.000000         -nan
7          -nan     0.000000         1097.045756   -nan     0.000000         -nan     710      0.000000         -nan

########################################################################################################################################################

I have tried the same but slightly modifying the moving points list ( decreasing 1 unit of y_direction ). Then I don't get the -nan output and the registration finishes (I haven't done enough tests to decide if it does what I need). I also attach the log file called elastix_modified_points.log

Btw, any idea why I get the -nan output? Shall I, instead of doing the "manual" registration, initialize the b-splines registration with the y translation?

Albert
2011/7/25 <M.Staring at lumc.nl<mailto:M.Staring at lumc.nl>>
Hi Albert,

It is available from elastix version 4.4, so if you are running that version you are good.
Parameters should be similar to the attached file.

Regards,

Marius

From: krekbert at gmail.com<mailto:krekbert at gmail.com> [mailto:krekbert at gmail.com<mailto:krekbert at gmail.com>] On Behalf Of Albert Gubern Mérida
Sent: maandag 25 juli 2011 16:21
To: Staring, M. (LKEB)
Cc: insight-users at itk.org<mailto:insight-users at itk.org>
Subject: Re: [Insight-users] B-Splines registration with fixed points

Hi Marius,

Thanks a lot. It seems to be what I exactly need. Anyway, is this registration available to run using the binary elastix (what I have been using at the moment)? If not, which components or/and testing files should I use to implement it?

Thanks again, and sorry for my ignorance.

Albert
2011/7/25 <M.Staring at lumc.nl<mailto:M.Staring at lumc.nl>>
Hi Albert,

If you have the coordinates of the sternum in both fixed and moving image you can add a term to the registration framework that penalizes deviation away from alignment of that point (or multiple points). We have done something similar:

http://elastix.isi.uu.nl/marius/publications/2011_c_MICCAIb.html

code of which can be found in elastix, which you can use freely.

But it all depends on whether you have corresponding pointsets.

Marius

From: insight-users-bounces at itk.org<mailto:insight-users-bounces at itk.org> [mailto:insight-users-bounces at itk.org<mailto:insight-users-bounces at itk.org>] On Behalf Of Albert Gubern Mérida
Sent: maandag 25 juli 2011 15:41
To: insight-users at itk.org<mailto:insight-users at itk.org>
Subject: [Insight-users] B-Splines registration with fixed points

Hi all,

I am currently working in Breast MRI registration between different patients, especially interested in the alignment of body area (thorax + pectoral muscles). The first step of my registration consist of a translation to align the sternum, which gives a good initial alignment of the pectoral muscles. In the following link you can see an example of one slice with the sternum indicated with a white cross: http://imageshack.us/photo/my-images/220/breastmristernum.png/. After the translation step with the pectorals already "rigidly" aligned I apply non-rigid b-splines registration to the whole volumes. It works well in some cases, but in others, especially of course when the differences are high between fixed and moving images, the non-rigid registration move the sternum point and the pectorals are not aligned anymore.

I have thought that not allowing the non-registration the move of the sternum point and the ones close to it could be a good solution for such problem. However, I have taken a look to the classes and the mailing list but I have not found any direct/easy example to how to do that.

Could anyone give me some advice? Does a easy way to do it exist or I should modify the classes involved in the registration framework? Which ones should I take a look?

Thanks in advance,

Albert

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20110726/831ec2d8/attachment.htm>