Registration¶
The Registration module provides several algorithms to align two or more datasets, such as images, in a common coordinate system.
In the ImFusion Suite, all datasets have a position relative to the global center at (0, 0, 0). However, this does not mean that the anatomical structures of two datasets are aligned. In the following example, the ultrasound is slightly shifted to the right compared to the MRI. With the help of the Image Registration algorithm, the images can be aligned.
Before registration |
After registration |
This module offers different algorithms for this kind of image alignment. After selecting two datasets in the ImFusion Suite, those algorithms are available in the context menu under the Registration category.
A registration will generally only change the position of one of the input datasets. This is usually referred to as the “moving” dataset, while the other dataset is referred to as “reference” or “fixed”. In general the dataset selected second is used as the moving dataset.
Image registration can be achieved by two means: the algorithm can change the global pose matrix of the dataset and optionally also attach a local per-pixel non-linear deformation. Every dataset has a matrix property, that describes the position of the dataset relative to the global coordinate system. The matrix can be inspected and modified with the Edit Transformation algorithm. The UI of this algorithm also allows to save and restore transformations, so that the can be compared easily.
Unlike the pose matrix, a deformation (a non-linear transformation) is fully optional and must by explicitly attached to a dataset either manually or as side-effect of an algorithm. An active deformation is indicated by a small icon next to the dataset:
The properties and type of the deformation can be inspected with the Edit Deformations algorithm. In general we recommend to apply (“burn-in”) a deformation after you are satisfied with it, since a lot of algorithms and annotation ignore deformations.
-
Aligns two images solely according to their pixel intensities. The algorithm provides different methods of comparing pixel intensities, as well as different transformation models. A Rigid transformation only changes the translation and rotation of the moving dataset, while an Affine one additionally changes the scale and shear. The different non-linear models will add a deformation to the moving dataset.
-
Aligns two images by manually placing points on corresponding landmarks in both images.
-
Aligns two point clouds or meshes.