[vtkusers] vtkDICOMReader and vtkGDCMImageReader orientation

[David Gobbi]

On Mon, Apr 4, 2016 at 4:09 AM, Richard Brown <richard.j.brown at live.co.uk>
wrote:

>
> I read that scaling, reflection/rotation, shear and translation are
> included in the 4x4 matrix. Given that shear is included in this matrix,
> shouldn’t gantry tilt be accounted for when I use GetPatientMatrix()?
>

Yes, the shear will be will be included in the matrix.  However, the
marching cubes algorithm expects a rectangular volume and will produce
incorrect results if given a sheared volume (unless the shear is very
small).  Actually, pretty much any image processing algorithm that looks at
voxel neighbors will give incorrect results if the degree of shear is such
that the neighbors found by connectivity (i.e. connected by faces) are not
the nearest voxels in the spatial sense (i.e. by center-to-center distance).

Imagine a CT voxel that is 1mm x 1mm x 5mm.  Now imagine that the
slice-to-slice shear is greater than 1mm, and draw the resulting
parallelogram and its neighbors, and hopefully you'll see the problem.
Even basic operations like linear interpolation give poor results in such a
situation, because the neighbors by connectivity are not the nearest
neighbors spatially.

Hence, a gantry-titled CT should be rectified before any 3D processing is
applied to it.  It can be un-rectified (re-sheared) before saving it back
to disk.  The in-plane resolution of CT is generally very good (1mm or
better) so rectification and de-rectification has very little impact on
image quality.


And then moving on from 3D surfaces…
>
> If I didn’t want to produce 3D surfaces; if I wanted to manipulate
> vtkImageData, is there any precautions I should take?
>
> For example, if I want to superpose DICOM RT-Structures or other surfaces
> (notably .STL in my case) onto my vtkImageData, do I need to apply any sort
> of matrix to either the DICOM vtkImageData or the superposed structures to
> make sure that they superpose in the correct position with the correct
> orientation?
>

My approach is to apply the matrix when I render the image data.  The
vtkImageActor, vtkImageSlice, vtkVolume etc. have a SetUserMatrix() method
that you can use to correctly orient the image within the world coordinate
system.  Generally, this means that I also have to correctly set the camera
position and focal point in order to get the desired view of the image.


And lastly, to get the world coordinates (x,y,z) from my vtkImageData
> coordinates (i,j,k), do I simply apply the GetPatientMatrix()? Is there a
> method for this? And the same question applies for world coordinate —>
> index coordinate conversions - I assume I calculate the inverse matrix and
> apply that in a similar fashion?


The vtkImageData (i,j,k) coords are called "structured coordinates" in
VTK-speak.  Applying the Origin and Spacing to these coords gives the "data
coordinates".  VTK's data processing filters work in data coordinates, and
if the spacing & origin are in millimeters, then the data coordinates are
also in millimeters.  The patient matrix must be applied to the data
coordinates, not to the structured coordinates.

Yes, there is probably a method to convert structured coords to data
coords, but I always just explicitly multiply by the spacing and add the
origin.

Cheers,
 - David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtkusers/attachments/20160404/26591b45/attachment.html>