Image registration is a key task in many image processing applications. In medical imaging, registration is necessary if two (or more) images have to be aligned in a common coordinate system such that corresponding image content is spatially coinciding.
Our registration techniques are based on a strong background in image registration. Therefore, we can provide customized high-tech registration solutions for various applications.
For example: motion compensation for post-processing, motion-corrected reconstruction, the alignment of different coordinate systems in image fusion or the compensation of organ movement.
Generally, we distinguish between rigid and non-rigid registration. In rigid registration, the transformation between the images is limited to only rotations and translations. This method is used to establish a link between different scanner coordinate systems or to account for rigid body movements.
In contrast, our non-rigid registration techniques allow far more degrees of freedom. It can be used to compute complex deformations between images, for example due to breathing motion and soft tissue deformations. Even inter-patient registration is possible in order to build atlases of various subjects.
The following list summarizes the main features of our registration techniques.
Rigid Registration
- Multi-modal input data (CT, MR, SPECT, PET)
- 2D-2D, 3D-3D, 2D-3D
- Data-independent and robust parameter estimation
- Model-to-image registration
- Fully automatic, landmark-based and manual registration
- Pre-processing for optimal registration results
- Intensity inhomogeneity correction (MR)
- Different field-of-views
- Volume masking (punching)
- Focus on arbitrary ROIs (regions of interest)
- Incorporation of intensity transfer functions
- GPU hardware acceleration
Non-Rigid Registration
- Multi-modal input data (CT, MR, SPECT, PET)
- 2D-2D, 3D-3D
- Fully automatic approach
- Computation of dense deformation field
- Incorporation of manual correspondences (landmarks, structures, ROIs)
- Statistical deformation models
- Simple parametrization with use cases
- Different field-of-views
- Volume masking (punching)
- Volume cropping
- GPU hardware acceleration
PET-CT Registration of Melanoma Patient
Compenstation of breathing motion in PET-CT image pair of melanoma patient. The images have been acquired at different times with single scanners. The example shows an exemplary slice from the 3-D volumes.
|
|
|
|---|---|---|
Initial image placement. The offset is due to the different scanner coordinate systems. | Result of automatic rigid registration, establishment of link between scanner coordinate systems. | Result of non-rigid registration. The breathing motion was corrected, along with motion at the heart. |
CT-MR Registration of Brain Tumor
Initial position of the two images:
|
|
|
|---|
Result of automatic rigid registration algorithm:
|
|
|
|---|
