Dermatology Image Annotation FHIR Implementation Guide - Local Development build (v0.1.0) built by the FHIR (HL7® FHIR® Standard) Build Tools. See the Directory of published versions

Terminology

Terminology

This page describes the terminology design decisions in this IG, including the use of standard code systems, custom CodeSystems, and binding strength choices.

SNOMED CT for clinical findings

Skin findings use SNOMED CT codes from the hierarchy under 106076001 | Skin finding |. SNOMED CT is the international standard for clinical terminology, and its dermatology hierarchy provides the granularity needed to represent diagnoses from melanoma to benign keratosis. The DermatologyFindings ValueSet uses an extensible binding, allowing implementations to use SNOMED codes outside the ValueSet when clinical need requires it.

SNOMED CT for body sites

Body site coding uses the BodySitesSkin ValueSet, which draws from SNOMED CT anatomical structure codes relevant to dermatology. This ValueSet also uses an extensible binding to accommodate site descriptions not covered by the initial value set.

LOINC for observation codes

The profiles use LOINC codes for Observation.code where appropriate, following the convention established by US Core and other FHIR IGs that LOINC is the standard system for clinical observation identification.

Extensible vs. required bindings

This IG applies binding strength based on the nature of the coded element:

• Extensible bindings are used for clinical codes (SNOMED findings, body sites) where implementations may encounter values outside the defined ValueSet. Senders SHOULD use a code from the ValueSet when one fits, but may use other codes when necessary.
• Required bindings are used for controlled vocabularies defined by this IG (annotation methods, QA status, change types) where the set of valid codes is closed and all implementations must use the same codes for interoperability.

Custom CodeSystems

This IG defines four CodeSystems:

• AnnotationType provides codes for Observation.code values (lesion-classification, segmentation, full-annotation, quick-classify), the dermatology-annotation category code, and all component slice discriminator codes (confidence-score, qa-review-status, inter-annotator-agreement, and others).
• AnnotationMethod defines the five methods by which an annotation can be produced, from expert manual annotation through pure AI inference. These codes implement AI Transparency requirements by making the level of AI involvement explicit and machine-readable.
• QAStatus provides quality assurance review status codes (approved, rejected, pending-review, revision-requested) for the QA review component.
• ChangeType defines change classification codes (unchanged, size-change, color-change, morphology-change, diagnosis-change, new-lesion, resolved) for the AnnotationSeries change detection component.

ISIC to SNOMED CT ConceptMap

The ISICtoSNOMED ConceptMap maps the eight ISIC diagnostic categories (MEL, BCC, SCC, NV, AK, BKL, DF, VASC) to their SNOMED CT equivalents. This mapping bridges the research community, which uses ISIC categories in published datasets and classification challenges, with clinical systems that require SNOMED-coded findings. Most mappings are equivalent; the BKL (Benign Keratosis) to Seborrheic keratosis mapping uses wider because ISIC's BKL category encompasses several benign keratosis subtypes.