The problem the set solves
The premise is stated in one sentence on the project site: “Standard identifiers for discrete anatomic locations would enable numerous levels of interoperability if applied broadly.”1 Two systems that agree a lesion sits at RID2772 agree on the kidney without sharing a database, a vendor, or a schema.
The obvious way to get such identifiers is to adopt an existing anatomy ontology whole. The site says why that was rejected. Attempts to use RadLex or SNOMED CT directly run into three problems:1
- Lack of numerous desired terms. Structures radiologists name every day are missing.
- Too many unnecessary or degenerate terms. Several concepts denote the same structure, so a coder has no principled way to choose one and a consumer cannot tell two records mean the same place.
- Limited anatomic organization. The hierarchy that exists does not answer the question imaging asks, which is what physically contains what.
Curation as the answer
The response is a curated subset rather than a new ontology: “We are curating a subset of anatomic concepts from existing ontologies and shaping them into a usable collection of anatomic identifiers for informatics interoperability.”1 The method was to start from RadLex terms already recognized in radiology reports, keep the used and useful ones, place each in an anatomic hierarchy, attach cross-references to other ontologies, and record laterality and sex phenotype.1
Two properties are claimed for the result and are the point of the exercise. It is complete enough, containing “almost all clinically used anatomic terms.” It is non-degenerate, with “no uncertainty as to which node represents a structure.”1 Non-degeneracy is what makes the identifier usable as a join key. See anatomic location for the term itself and the data model for the record shape.
Separable identity
Anatomy is one of the two axes of an Observation, and the model keeps it separate from the other. The January 2026 status update states the atomic unit as “what plus where plus attributes: finding tag plus anatomic location plus lesion characteristics,” and says anatomy is “baked into OIFM definitions and Observation objects.”2 The set is “curated to identify where findings are visualized on imaging exams,” which is a narrower target than anatomy in general and explains why structures with no imaging appearance are absent.2
Separating what from where is what lets a single finding model for a cyst serve the kidney, the liver, and the breast, and what lets the Imaging Problem List offer “precision filtering via anatomy-embedded definitions.”2
Where the identifiers are used
| Where | How anatomy appears |
|---|---|
| Finding model definitions | FindingModelFull.anatomic_locations, a list of index codes naming the locations a finding can occupy3 |
| Exam Finding List observations | An optional anatomicLocation object of locationId and locationDisplay, a single location per observation, added on the extraction platform’s development branch4 |
| Imaging Problem List grouping | Entries are grouped by the pair of finding code and location identifier rather than by finding code alone, so a left and a right lesion are separate problems4 |
| Finding and location coding | The coding pass generates search terms, searches the location index, and has a selector choose from the returned candidates, with assignment governed by a stated precedence ladder5 |
| Imaging Problem List viewer | Region and organ-cluster filters, and a body schematic whose zones are driven by the location’s region and laterality6 |
The rule that keeps this honest is stated with the assignment rules: every assigned identifier must exist in the set, and a structure absent from it is left unassigned rather than forced to a wrong code.5 That discipline turns missing terms into a visible list of requests rather than silent mis-coding. See RadLex integration for where those requests go.
What curation costs
A curated set is a maintained artifact, not a free one. The site’s own roadmap names the standing content work: review the hierarchy, add, prune, and modify; edit and improve synonyms; complete SNOMED identification.1 It also names the companion effort that has not been built, “a companion for exam types based on LOINC/RadLex Playbook exam definitions that specify all included body parts for the exam,” which remains the earliest written statement of the exam type goal.1 See exam types.
Footnotes
-
anatomiclocations.org site homepage, rationale, approach, and roadmap ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7
-
Open Imaging Data Model 2026 Status Update, January 2026 ↩ ↩2 ↩3
-
FindingModelFull definition, findingmodel repository, main ↩
-
Plan to add anatomic-location codes to Exam Finding Lists and an anatomy-aware Imaging Problem List ↩ ↩2
-
Anatomic location assignment rules, imaging-problem-list dev ↩ ↩2
-
Anatomy-aware viewer plan, imaging-problem-list dev ↩