[ITK-users] What is an accepted duration of a 3D registration? Should I expect the MSE to "converge" to 0?

[Siavash Khallaghi]

Marina Bendersky wrote
> 1- Would it be correct to normalize all images with the same filter
> (like NormalizeImageFilter), and then create the volumes and start the
> registration?

1. You could try using an  itk::HistogramMatchingFilter
<http://www.itk.org/Doxygen/html/classitk_1_1HistogramMatchingImageFilter.html>  
before doing registration. This /should/ ensure that both images occupy the
same dynamic range.

2. Alternatively, you can try using a metric that is less sensitive to
linear changes in intensities, e.g.  NCC
<http://www.itk.org/Doxygen/html/classitk_1_1NormalizedCorrelationImageToImageMetric.html> 
. I am assuming that you are doing mono-modality registration.


Marina Bendersky wrote
> 2- You're right, there are probably local deformations and I've only
> been trying rigid/affine transforms. I'm comparing two volumes of
> different time points and there are likely to be transformations. I
> understood, however, that a deformable transformation might change the
> images (ie. the lesions), so I completely ruled deformable
> transformations out because my objective is to assess how the lesion
> changes over time ...
> 
> Is this correct? Or, should I still try a deformable registration?

If your goal is to assess the geometry of the lesion (i.e. how much it has
grown/shrunk), then deformable registration is actually what you need. The
output of deformable registration is simply a deformation field that maps
spatial locations in the fixed volume into the moving. You can interpolate
the deformation field on your fiducials (the points on the surface of the
lesion) to calculate the amount of change in your lesion between the two
volumes.


Marina Bendersky wrote
> Using the ITK outputs (matrix, translation, and center values), I
> computed the new coordinates of the points and I hoped that the new
> "z" value will determine the "correct" slice of the moving volume
> (time 2). Does this sound correct to you?

Seems like you are tyring to use the registration transform to calculate the
target registration error using fiducials identified on your lesions. The
whole idea is correct and has been the gold standard in registration for
years. However, I am not sure whta you mean by the z-slice. Maybe my
previous comment is of help to you.

HTH,

Siavash



--
View this message in context: http://itk-users.7.n7.nabble.com/ITK-users-What-is-an-accepted-duration-of-a-3D-registration-Should-I-expect-the-MSE-to-converge-to-0-tp36133p36141.html
Sent from the ITK - Users mailing list archive at Nabble.com.