The AIM information model is a component of the National Cancer Institute’s (NCI) Cancer Biomedical Informatics Grid (caBIG) project. As such, the AIM model has received caBIG silver level compatibility certification. The latter means that concepts and relationships used in the model, as well as the model itself, have been harmonized with other caBIG modeling efforts to insure reuse of common data elements and controlled terminologies. Predominantly, over the past two years, AIM has been adopted in a variety of research settings. This paper describes changes, based on feedback from the community of adopters, to the AIM information model and tools that have resulted in the development of AIM version 2.0.

1. The type of annotation being made was previously defined as a list of controlled types (approximately 20). These types had a set of predefined values and could not be rapidly expanded without modification to the model. An AIM user could not temporarily create a new or private type without violating the model. In AIM version 2.0, we replace the type attribute with references to controlled terminologies, such as DICOM, SNOMED, or others. The controlled terminology reference is a triple that specifies the terminology in question, the code of the term, and the code meaning of the term. In this manner, the types of allowed annotations can be more easily extended, simply by extending the definitions in a controlled terminology. This also supports the use of private or temporary vocabularies.

2. Some DICOM-specific metadata has been removed from the AIM model. The DICOM model is very rich in metadata about how the image was acquired and its technical parameters. Since, technically, such meta data is not a component of the annotation (but rather of the image), and since it is practically impossible to select a priori (which information is to be included), it was decided to remove DICOM meta data from the AIM model, with the exception of the DICOM-unique identifiers (UIDs) by which reference to the image is made.

3. The AIM model previously supported the concept of DICOM probability maps related to an imaging observation. In this way, the probability that a set of pixels has a certain meaning or the fractional composition of pixels may be represented. The probability map has been replaced with a reference to formal DICOM segmentation objects, either binary or fractional.

4. In order to use existing caDSR common data elements (CDEs) and to be in compliance with the National Biomedical Imaging Archive (NBIA) model, classes in AIM model version 2.0 have been renamed from Study, Series, and Patient to ImageStudy, ImageSeries, and Person, respectively.

3. Two new classes, AnatomicEntityCharacteristic and Rating, have been added to the model. AnatomicEntityCharacteristic are characteristics of anatomic entities. These are in contradistinction to ImagingObservationCharacteristic, which are related to the observation. For example, “spiculated” is an ImagingObservationCharacteristic of the Observation, “mass,” while “dilated” might be an AnatomicEntityCharacteristic of the AnatomicEntity, “colon.”

Rating can be used to quantify or grade a concept (e.g. , “severity”) with a numerical value associated with the concept. Ratings are associated with both AnatomicEntityCharacteristic and ImagingObservationCharacteristic classes.

4. As was done with annotation type above, calculation types were previously modeled as a controlled list of choices. In AIM version 2.0, the type of calculation is now defined by reference to a controlled terminology. Again, this allows for more facile extension of these definitions.
A calculation can now be directly associated with a graphical markup. The association is made through the ReferenceGeometricShape class. The class has an attribute that captures a unique integer number that is assigned to a graphical shape. Calculation results, such as distance, area, and maximum and minimum pixel value, become directly linked to a graphical region. This enables the use case of multiple regions of interest each possibly with different calculations yet defining a single annotation.

5. Annotations may have different roles in a particular study or a clinical trial. In particular, it is important when creating Annotation of Annotations, to specify what role each referenced annotation plays in that particular analysis. The AnnotationRole class is created to describe the role of referenced annotation. A role is defined via controlled terminology and annotation’s sequence number within the role.

6. The Inference class was added to capture a conclusion derived by interpreting imaging observations and their characteristics. Many “findings” made from imaging observations are, in fact, inferences from those observations. Inferences are applied to the entire image annotation or an annotation of annotation.