The basic use case for the transformation from AIM to DICOM SR is exchange of quantitative and categorical information about DICOM images and regions of interest in DICOM images.
AIM and DICOM SR instances both contain references to DICOM images, and both make use of references to other types of DICOM instances such as segmentations. The flow of information between systems that might make use of a conversion between AIM and DICOM SR instances is considered in Figure A.2-1.
Various different transformation scenarios should be considered:
-
Transformation of a complete AIM instance to a complete DICOM SR Measurement Report. The receiver optionally selects relevant parts of the transformed document for further processing or static or interactive visualization by the user.
-
Transformation of a subset of a AIM instance to a DICOM SR Measurement Report. This subset comprises the relevant information for a specific use-case.
-
Transformation of multiple related AIM instances to a single DICOM SR Measurement Report. The selected related AIM instances comprise the relevant information for a specific use-case.
This Part of the Standard does not mandate any particular transformation scenario. Transformations of various compatible components of AIM and DICOM SR TID 1500 are described, allowing each of these scenarios to be implemented as appropriate. This Part of the Standard enables a deterministic transformation of the first scenario (complete mapping of an AIM instance to DICOM).
The primary use-case for the transformation is that of:
-
taking an existing AIM instance containing one or more annotations, converting it into DICOM SR, storing it in a DICOM storage system, possibly displaying it with a DICOM SR aware application, transforming it back into an AIM instance after retrieval from the DICOM storage system, and reusing it in an AIM-aware application
An important secondary use-case is transformation of annotation information in different formats into a single format for data aggregation (e.g., "analytics", "data mining", "big data", "machine learning", and "deep learning"). The conversion described here is intended to allow preservation of semantics sufficient for such purposes, regardless of the source format.
It should be understood that DICOM SR created by transforming an AIM object will not necessarily be identical in structure and content to a DICOM SR that might have been created de novo by a similar application. For instance, there are various encoding choices that an application implementer may make, especially with respect to the degree of post-coordination of ROI and measurement descriptions, which might result in different structures. It may not be possible to transform a DICOM SR TID 1500 instance into AIM and retain all of its content. Post-coordinated concepts, such as measurement and derivation methods, may be preserved by using multiple CalculationEntity/typeCode entries or by pre-coordinating into a single concept during the transformation.
Multiple regions of interest, or multiple measurements and categorical statements about a single region of interest, or about the same real world entity (e.g., lesion) identified with different regions of interest (e.g., at different time points or with different modalities), may be encoded in single or multiple AIM or DICOM SR instances. Whether or not a source AIM or DICOM SR implementation encodes more than one region of interest (and their accompanying measurements and categorical statements) in a single instance or in separate instances, and whether the conversion from one form to the other "bundles" multiple instances into a single instance, or "unbundles" a single instance into multiple instances, is not prescribed. The AIM 4.2 model allows for the encoding of multiple marked up regions and multiple measurements in a single annotation instance, and unlike earlier AIM model versions provides mechanisms for identifying which markup is associated with which measurement using AIM statements; otherwise it is necessary to assume that all markup applies to all measurements and vice versa.
For example, for RECIST measurements that involve the long and short axis of a lesion, it is not only possible to encode in AIM (and map to TID 1500) the boundaries of and the area measurements derived from a planar volumetric ROI, it is also possible to encode the endpoints of the measured axes of the pair of linear measurements and their derived values.