The next-generation schema
Both Open Imaging Finding Models and ACR/RSNA Common Data Elements are moving onto one schema, being developed in the ACR/RSNA CDE schema repository on its next-gen-2026 branch, as a graph of interconnected concepts rather than a set of documents. Applications will be able to use either or both collections at once via a common SDK based on the new schema. What follows is the design as it currently stands; it is still being worked out.
(click the image for full size)
Node types
| Node type | Defines | Example |
|---|---|---|
| FindingClass | the properties of the items described in imaging reports | pulmonary nodule; part-solid pulmonary nodule |
| Diagnosis | a separate, closely related node type, distinguished by a different set of possible relationships | lung cancer |
| Grouping | broad classes of findings/diagnoses, usually intended to be used in the negative | pulmonary parenchymal abnormality; renal abnormality |
| DataElement | a categorical descriptor: a value set, ordered or unordered, whose values are themselves nodes | margin: smooth, lobulated, irregular, spiculated |
| Measurement | a quantitative descriptor by the kind of quantity, such as length or CT density, with the unit family following from the kind | mean diameter |
| AssessmentScheme | the systems via which a constellation of other findings is evaluated | Lung-RADS |
| AnatomicLocation | the entries of the anatomic locations index, keyed by their RadLex identity | lung |
Edges that build a definition
Edges integrate the nodes into a rich fabric of interrelated concepts, enabling reliable, reproducible inference. Some edges build a single definition, binding it to the descriptors that specify its attributes, to its anatomic scope, and to its standard clinical metadata:
| Edge | Binds a definition to | Example |
|---|---|---|
| HAS_DATA_ELEMENT | a DataElement that characterizes it | pulmonary nodule → margin |
| HAS_MEASUREMENT | a Measurement that quantifies it | pulmonary nodule → mean diameter |
| SCOPED_TO | its anatomic scope | pulmonary nodule → lung |
| SEEN_ON, IN_SUBSPECIALTY, HAS_ETIOLOGY, and the other metadata edges | its standard clinical metadata (below) | pulmonary nodule → CT |
The edges themselves may contain properties that provide additional detail: a binding may, for example, restrict a shared DataElement to a subset of its values for one FindingClass without changing the element itself. No element is required by any definition.
Other edges relate definitions to one another: subtypes, components, manifestation, causation, progression, co-occurrence, and assessment. Those are described in Relationships among definitions.
Anatomic scope
A definition’s anatomic scope says where that kind of finding belongs. The scope may be a place, such as the kidney, or a kind of structure, such as muscle. It is an edge, SCOPED_TO, from the definition to an anatomic location. Descriptors carry scope too: a margin element for pulmonary lesions is scoped to the lung.
Scope is guidance, not a location. The definition of a clavicle fracture is scoped to the clavicle; an actual fracture, described on a patient’s exam, is at the left clavicle. The description of a specific finding points to its definition and, beside that, to its own location, and the location must fall within the definition’s scope: a left kidney is a kidney, so it satisfies a scope of “kidney”. Location is not one of a definition’s attributes, so there is no location DataElement.
Laterality is part of the location, not a separate descriptor. Left, right, and unsided versions of a lateralized structure are distinct anatomic locations, so there is no laterality DataElement either.
A scope can name a tissue type, such as lung parenchyma, a region, such as the lung, or both. A pulmonary nodule is scoped to both: the tissue type lung parenchyma and the region lung. A specific nodule’s location then names the region, the right upper lobe for example, and the tissue type comes from the definition without being restated.
The anatomy itself, the curated index of locations anchored in RadLex, is described in Anatomic locations.
Standard clinical metadata
A definition can carry seven standard facts beyond its attributes, each drawn from a shared, closed list of values:
| Fact | What it says about a definition | Example values |
|---|---|---|
| Modality | the modalities on which it is visible | CT, MR, US, XR, PET, NM |
| Subspecialty | the subspecialty it belongs to, for filtering and routing | chest radiology |
| Body region | the region of the body; derived from the anatomic scope rather than stated separately | thorax |
| Sex | whether it is specific to one sex | sex-neutral, female-specific, male-specific |
| Age | the ages to which it applies | all ages, newborn, … |
| Time course | its typical duration and course | hours to permanent; progressive, stable, resolving, recurrent |
| Etiology | for a diagnosis, the broad kind of cause | infectious, neoplastic, traumatic |
A missing value means no information, not “none”.
Evolution of CDEs and Finding Models
Both CDEs and existing Open Imaging Finding Models will have to migrate from their current JSON document-based organization to a new graph-oriented format, likely using RDF/OWL as an underlying storage format. More ergonomic data structures will also be developed for use in SDKs and via an API.