SetForceOrthogonalDirection not having effect

Beginner Questions
1

Hi,

I got a .dcm serie of images with gantry tilt. I loaded them with ITK5.1 and ITK4.10, but both of them give a skewed image. Using Dicom2Nifti doesn’t have this problem. Photo as example:

image
image.png736×316 112 KB

This is the code I use:

auto volumeReader = itk::ImageSeriesReader<InputShortImageType>::New();
volumeReader->SetForceOrthogonalDirection(false); // Doesn't solve gantry tilt problem
auto movinggdcmIO = itk::GDCMImageIO::New();
volumeReader->SetImageIO(movinggdcmIO);
auto volumeInputNamesGenerator = itk::GDCMSeriesFileNames::New();
volumeInputNamesGenerator->SetInputDirectory(inputDirectory);
volumeReader->SetFileNames(volumeInputNamesGenerator->GetInputFileNames());
volumeReader->Update();

Is there anything I am overlooking?

2

Have you tried volumeReader->ForceOrthogonalDirectionOff();?

3

Also, you are viewing the image using ITK-SNAP, which ignores non-orthogonal basis vectors. Try visualizing the images using 3D Slicer, which has better support for it.

4

Hey @dzenanz, thanks for the swift reply!
Interesting enough, it indeed does work when exporting to .nrrd and visualised with 3d-Slicer
:o However, all other ways are skewed. (.nii with 3d-slicer is skewed, .nrrd with ITK-Snap is skewed, .nrrd with an older VTK)
It look like the meta date of the .dcm files is passed trough to the .nrrd. Maybe, is there a way to resample the image to an orthogonal axis system?

5

Of course! By the looks of it, that’s what Dicom2Nifti does. @bpaniagua is the tool from craniosynostosis project publicly available? If not, translating dicom2nifti resample into ITK is not trivial, but should not be hard.

6

If you want to have image with correct geometry and orthogonal axes then you need to resample the image.

You can do this in 3D Slicer by enabling regularization (application settings / DICOM / DICOMScalarVolumePlugin / Acquisition geometry regularization → apply regularization transform) and resampling the volume by Crop volume module (or other resampling modules).

I would not recommend porting dicom2nifti’s simplistic approach to ITK, as it can only deal with tilted-gantry acquisition, but fails for other non-regularities, such as non-uniform slice spacing. 3D Slicer’s implementation builds a grid transform from all transformed slice corner points, which is simpler, more elegant, and it correctly handles all kinds of non-regularities.

1 Like