Page MenuHomePhabricator

Image orientation flip upon segmentation start (Siemens Aera 1.5T MRI Scanner, DICOM imaging data)
Closed, DuplicatePublic

Description

Hi everybody,

thanks for MITK, which is a great tool!
We are evaluating to employ MITK in a MRI project, we face a problem regarding data loading and segmentation:

Initially, loading the DICOM series from a local directory works and gives us correct orientation (meaning axial, coronal, sagittal) in the viewer. But upon starting the segmentation, the orientations get flipped. This happens when I create a new segmentation, after entering its name. Now the window named "axial" shows a coronal reconstruction, while "coronal" shows the axial images and "Sagittal" gives a prone view head left.

A "Reinit" on the image series does not help, while a "Global Reinit" restores correct image orientation, but necessitates a reinit on the segmentation, which again flips the images as noted above.

I figure this might be related to http://mitk-users.narkive.com/SvdIsLSz/question-reinit-image-flip, but the related bugs show as fixed. We're using MITK Workbench 16.03, are those fixes included?

Imaging data comes from a clinical MRI scanner (Siemens Aera 1.5 T), the sequence is a T1 Dixon 3D sequence. Another type of Sequence (2D T2 imaging) does not seem to exhibit this problem. Anonymized dicom imaging data can be provided.

Thanks a lot for any help,

Joost

Event Timeline

The mentioned fix for T11477 was merged in January of 2016 (rMITK8e0b85)and is part of the 2016.03 release, T16772 was a duplicate of that task.
However these fixes only prevent the planes from being unusable if the handedness is unexpected.

Doublicate of T19632, working on it, seems to be a nasty problem.

We need to check the reinit()-functions, looking for things that aren't there but ougth to be, like distinction of cases left/righthandedness, explicit geometry (RAS, LPS, RPS, whatever) or guessing.

Currently the labels "Axial", "Coronal" and "Sagital" are never changed, which is, ahem, optimistic.

Cheapest "solution" would be to reject data with det( index2worldM ) < 0, but obviously not practical, since scanners sometimes give us lefthanded geometries.

Thanks everybody for taking this up.

Is there a need for exemplary imaging?