Registered

---

Registration

Registration is the process by which two medical images taken at different times, corresponding to the same part of the body and generally to the same patient, are overlaid coincidentally in order to check the evolution of an injury/disease/operation.

By means of this image comparison, the differences between the two are studied, but in order to make this possible in the most intuitive way possible, the overlapping coincidence must be achieved as precisely as possible. This has to be done with two images that, in addition to being taken at different times, may come from different equipment (another scanner model, for example) and where it cannot be guaranteed that the position in which the patient was placed was exactly the same (variations in orientation and displacement).

A summary of the possible registration scenarios is:

• quasi-matching anatomical structures: two images in which the same anatomical structures, of similar size, are present in both.
• Variations in patient size: particularly in the case of children, where natural growth causes differences in whole body size.
• anatomical structures with different volumes: when the evolution of a tumour mass that has grown over time is to be studied, it has increased in volume in one of the images (it may be the most recent or the oldest) and may have displaced other anatomical structures or adjacent organs.
• non-existent anatomical structures: after surgery (or treatment) in which part of the tissue (organ, tumour, bone, etc.) has been removed, pairs of elements that do not exist in one of the images must be compared.

Description of the procedure

In general, the procedure to follow is to load both images one next to the other, identify common points in each one (the points are paired, one is taken in image A and the other in image B). If both images have the same orientation, no distortion will be necessary to match them, but this is not usually the case. Then you have to select which of the two images you are working with (A or B) will apply a distortion to achieve full coincidence.

When determining the common points in both images, it may be necessary to move between the different slices of the three available views so that the same anatomical information is displayed in the image A and B.

Modules

The registration modules can be found in the lower half of the module dropdown (Figure 1), they are grouped under the nameRegistration:

Figure 1: registration modules

Traditionally, brain imaging has been used, which is why several of the modules are oriented to these specific cases.

Terminology

• Landmark: representative points of the image (selected by the user in both images) to ensure coincidence
• Fixed volume: image that is not going to undergo any transformation
• Moving volume: image to be transformed to match the one selected asfixed volume.

Example: landmark registration

Volume loading

Once the two sets of images are loaded, the Registration\Landmark registration module is opened. At that moment the module asks the user to determine which is going to be thefixed volume and themoving volume (Figure 2) using the dropdown lists, when clicking on each of them you can see which are the volumes loaded in the current scene and select the corresponding one.

Figure 2: fixed and moving volumes

The screen layout changes to a 3x3 grid (Figure 3). By columns, the axial, coronal and sagittal views are shown in columns 1, 2 and 3 (this is the default layout, unless the user decides to modify it). By rows thefixed volume in row 1 is displayed, themoving volume in row 2 and the result of the overlapping of the two in row 3, which will be renamed-transformed.

Figure 3: registration layout

Select layout=

If you wish to modify this layout, it will be done from the module panel, modifying the following options (Figure 4). If any of the views are not relevant, you can deactivate them and thus take up no space on the screen, control the transparency of the image overlay and even the zoom:

Figure 4: layout

Also, if you want to change the volumes you are working with, a little higher up on the module panel, using the button'Select Volumes to Register you will return to the selection menu described in Figure 2.

For example, by selecting Coronal the layout remains like this (Figure 5):

Figure 5: coronal layout

Landmarks

The matching points, or landmarks, are added to the Landmarks section of the module panel, located just below the ones described above (Figure 6).

Figure 6: coronal layout

Press the ADD button and place the point (the cursor changes automatically and we are in fiducial mode) on thefixed volume by clicking on the desired place. The program will automatically place the same point (called L-0 in the example image) in themoving volume. Ideally, these points should be placed on elements that are representative of both images and that are easily identifiable.

Now, as shown in the image (Figure 7), the user has put the point in the view of the first column (green) and Slicer has put the point in the view of the second column (yellow). The landmark of the moving volume is not in the same place as the image of the fixed volume (Slicer simply places it in the same coordinates, without taking into account the image and what it represents), so to correct its location click on the REFINE LANDMARK button that is just below the list of landmarks present in the current registry. Then we can move manually, by clicking and dragging, the landmark of the moving volume to the position we consider visually correct.

Figure 7: landmak point

We continue adding landmarks until we consider that we have put the necessary ones for the correct adjustment of both images, at this point of the work process we will have a set of points (Figure 8), which are defined by their name. They can be selected by clicking on the cursor with the crosshair icon and deleted with the trash can icon.

Figure 8: landmak list

The refining of landmarks can be based on two refining methods, BRAINSFit and SimpleITK, but neither of these methods is a substitute for the operator and his criteria.

Registered

There is no canon that determines the method of registration to be applied, since it depends on different factors in each case, such as the type of images you are working with, their quality, the anatomy you are visualizing and the expected results. A standard method and quite open to configuration by the user is theAffine Registration (Figure 9), which allows these three levels of adjustment:

• Rigid: allows translations and rotations
• Similarity: rigid + uniform scale changes
• Affine: Similarity + shear

The affine transformation is also called a linear transformation and is defined by different matrices to pass from a set of initial coordinates to others [Wikpedia Affine Transformation].

Figure 9: linear registration

Depending on the quality and quantity of selected landmarks, the registered will be more or less accurate and it is the operator's task to refine the landmarks so that the result is as optimal as possible.

---

MENU