Annex O Registration (Informative)


These SOP Classes allow describing spatial relationships between sets of images. Each instance can describe any number of registrations as shown in Figure O.1-1. It may also reference prior registration instances that contribute to the creation of the registrations in the instance.

A Reference Coordinate System (RCS) is a spatial Frame of Reference described by the DICOM Frame of Reference Module. The chosen Frame of Reference of the Registration SOP Instance may be the same as one or more of the Referenced SOP Instances. In this case, the Frame of Reference UID (0020,0052) is the same, as shown by the Registered RCS in the figure. The registration information is a sequence of spatial transformations, potentially including deformation information. The composite of the specified spatial transformations defines the complete transformation from one RCS to the other.

Image instances may have no DICOM Frame of Reference, in which case the registration is to that single image (or frame, in the case of a multi-frame image). The Spatial Registration IOD may also be used to establish a coordinate system for an image that has no defined Frame of Reference. To do this, the center of the top left pixel of the source image is treated as being located at (0, 0, 0). Offsets from the first pixel are computed using the resolution specified in the Source IOD. Multiplying that coordinate by the Transformation matrix gives the patient coordinate in the new Frame of Reference.

A special case is an atlas. DICOM has defined Well-Known Frame of Reference UIDs for several common atlases. There is not necessarily image data associated with an atlas.

When using the Spatial Registration or Deformable Registration SOP Classes there are two types of coordinate systems. The coordinate system of the referenced data is the Source RCS . The coordinate system established by the SOP instance is the Registered RCS .

The sense of the direction of transformation differs between the Spatial Registration SOP Class and the Deformable Spatial Registration SOP Class. The Spatial Registration SOP Class specifies a transformation that maps Source coordinates, in the Source RCS, to Registered coordinates, in the Registered RCS. The Deformable Spatial Registration SOP Class specifies transformations that map Registered coordinates, in the Registered RCS, to coordinates in the Source RCS.

The Spatial Fiducials SOP Class stores spatial fiducials as implicit registration information.


Figure O.1-1 Registration of Image SOP Instances

O.2 FUnctional Use Cases

Multi-Modality Fusion: A workstation or modality performs a registration of images from independent acquisition modalities—PET, CT, MR, NM, and US—from multiple series. The workstation stores the registration data for subsequent visualization and image processing. Such visualization may include side-by-side synchronized display, or overlay (fusion) of one modality image on the display of another. The processes for such fusion are beyond the scope of the Standard. The workstation may also create and store a ready-for-display fused image, which references both the source image instances and the registration instance that describes their alignment.

Prior Study Fusion: Using post processing or a manual process, a workstation creates a spatial object registration of the current Study’s Series from prior Studies for comparative evaluation.

Atlas Mapping: A workstation or a CAD device specifies fiducials of anatomical features in the brain such as the anterior commissure, posterior commissure, and points that define the hemispheric fissure plane. The system stores this information in the Spatial Fiducials SOP Instance. Subsequent retrieval of the fiducials enables a device or workstation to register the patient images to a functional or anatomical atlas, presenting the atlas information as overlays.

CAD: A CAD device creates fiducials of features during the course of the analysis. It stores the locations of the fiducials for future analysis in another imaging procedure. In the subsequent CAD procedure, the CAD device performs a new analysis on the new data set. As before, it creates comparable fiducials, which it may store in a Spatial Fiducials SOP Instance. The CAD device then performs additional analysis by registering the images of the current exam to the prior exam. It does so by correlating the fiducials of the prior and current exam. The CAD device may store the registration in Registration SOP Instance.

Adaptive Radiotherapy: A CT Scan is taken to account for variations in patient position prior to radiation therapy. A workstation performs the registration of the most recent image data to the prior data, corrects the plan, and stores the registration and revised plan.

Image Stitching: An acquisition device captures multiple images, e.g. DX images down a limb. A user identifies fiducials on each of the images. The system stores these in one or more Fiducial SOP Instances. Then the images are “stitched” together algorithmically by means that utilize the Fiducial SOP Instances as input. The result is a single image and optionally a Registration SOP Instance that indicates how the original images can be transformed to a location on the final image.

O.3 System Interaction

Figure O.3-1 shows the system interaction of storage operations for a registration of MR and CT using the Spatial Registration SOP Class. The Image Plane Module attributes of the CT Series specify the spatial mapping to the RCS of its DICOM Frame of Reference.


Figure O.3-1 Stored Registration System Interaction

The receiver of the Registration SOP Instance may use the spatial transformation to display or process the referenced image data in a common coordinate system. This enables interactive display in 3D during interpretation or planning, tissue classification, quantification, or Computer Aided Detection. Figure O.3-2 shows a typical interaction scenario.


Figure O.3-2 Interaction Scenario

In the case of coupled acquisition modalities, one acquisition device may know the spatial relationship of its image data relative to the other. The acquisition device may use the Registration SOP Class to specify the relationship of modality B images to modality A images as shown below in Figure O.3-3. In the most direct case, the data of both modalities are in the same DICOM Frame of Reference for each SOP Class Instance.


Figure O.3-3 Coupled Modalities

A Spatial Registration instance consists of one or more instances of a Registration. Each Registration specifies a transformation from the RCS of the Referenced Image Set, to the RCS of this Spatial Registration instance (see PS 3.3) identified by the Frame of Reference UID (0020,0052).


Figure O.4-1 shows an information model of a Spatial Registration to illustrate the relationship of the attributes to the objects of the model. The DICOM attributes that describe each object are adjacent to the object.


Figure O.4-1 Spatial Registration Encoding

Figure O.4-2 shows an information model of a Deformable Spatial Registration to illustrate the relationship of the attributes to the objects of the model. The DICOM attributes that describe each object are adjacent to the object.


Figure O.4-2 Deformable Spatial Registration Encoding

Figure O.4-3 shows a Spatial Fiducials information model to illustrate the relationship of the attributes to the objects of the model. The DICOM attributes that describe each object are adjacent to the object.


Figure O.4-3 Spatial Fiducials Encoding


A 4x4 homogeneous transformation matrix describes spatial rotation, translation, scale changes and affine transformations that register referenced images to the Registration IE’s RCS. These steps are expressible in a single matrix, or as a sequence of multiple independent rotations, translations, or scaling, each expressed in a separate matrix. Normally, registrations are rigid body, involving only rotation and translation. Changes in scale or affine transformations occur in atlas registration or to correct minor mismatches.


Fiducials are image-derived reference markers of location, orientation, or scale. These may be labeled points or collections of points in a data volume that specify a shape. Most commonly, fiducials are individual points.

Correlated fiducials of separate image sets may serve as inputs to a registration process to estimate the spatial registration between similar objects in the images. The correlation may, or may not, be expressed in the fiducial identifiers. A fiducial identifier may be an arbitrary number or text string to uniquely identify each fiducial from others in the set. In this case, fiducial correlation relies on operator recognition and control.

Alternatively, coded concepts may identify the acquired fiducials so that systems can automatically correlate them. Examples of such coded concepts are points of a stereotactic frame, prosthesis points, or well-resolved anatomical landmarks such as bicuspid tips. Such codes could be established and used locally by a department, over a wider area by a society or research study coordinator, or from a standardized set.

The table below shows each case of identifier encoding. A and B represent two independent registrations: one to some image set A, and the other to image set B.

Fiducial Identifier (0070,0310) Fiducial Identifier Code Sequence (0070,0311)
Uncorrelated A: 1, 2, 3 B: 4, 5, 6 A: (1, 99_A_CSD, label A1) … B: (4, 99_B_CSD, label B4) …
Correlated A: 1, 2, 3 … B: 1, 2, 3 … A: (1, 99_MY_CSD, label 1) … B: (1, 99_MY_CSD, label 1)…

Fiducials may be a point or some other shape. For example, three or more arbitrarily chosen points might designate the inter-hemispheric plane for the registration of head images. Many arbitrarily chosen points may identify a surface such as the inside of the skull.

A fiducial also has a Fiducial UID. This UID identifies the creation of the fiducial and allows other SOP Instances to reference the fiducial assignment.