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

Examples

Examples

This page walks through each example resource in the IG, explaining the clinical scenario and the key elements demonstrated. All examples use synthetic patient data.

ExampleMelanomaAnnotation

Scenario: Dr. Robert Johnson, a board-certified dermatologist, reviews a clinical photograph of a lesion on the right upper arm of patient Jane Smith. He performs a full expert annotation, classifying the lesion as malignant melanoma with high confidence. The annotation passes QA review and achieves strong inter-annotator agreement.

Key elements:

• method is set to expert-annotation, indicating a specialist performed the full annotation without AI assistance
• component[skinFinding] uses SNOMED code 372244006 | Malignant melanoma |
• component[confidenceScore] records 0.92, the annotator's self-reported confidence
• component[qaReviewStatus] is approved, indicating the annotation passed quality review
• component[interAnnotatorAgreement] records a kappa of 0.87, reflecting strong agreement
• derivedFrom references the source clinical photograph (DocumentReference)

See ExampleMelanomaAnnotation.

ExampleAIPrediction

Scenario: Before Dr. Johnson reviews the image, the DermAI Classifier v2.1 analyzes it and produces a preliminary prediction of malignant melanoma with 89% confidence.

Key elements:

• meta.security includes the AIAST code, marking this as AI-generated content
• status is preliminary because no clinician has reviewed the prediction
• method is fixed to ai-inference
• device references the AI model Device resource (DermAI Classifier v2.1)
• component[predictionConfidence] records 0.89
• component[inferenceTimeMs] records 245 ms of model inference time

See ExampleAIPrediction.

ExampleAIPredictionProvenance

Scenario: A Provenance resource documents the audit trail for the AI prediction, recording which system created it and from what source data.

Key elements:

• target references the ExampleAIPrediction Observation
• agent.who references the AI model Device, identifying the system that produced the prediction
• entity.what references the source clinical photograph with role source

See ExampleAIPredictionProvenance.

ExampleAIAssistedConfirmed

Scenario: Dr. Johnson reviews the AI prediction and agrees with the malignant melanoma classification. He confirms the AI result without modification using the quick classify workflow, producing a final annotation.

Key elements:

• method is ai-assisted-confirmed, indicating the clinician accepted the AI prediction
• meta.security retains the AIAST code because the result originated from AI analysis
• status is final because a clinician has reviewed and accepted the finding
• code is quick-classify, representing the rapid clinician-driven classification workflow
• performer references the confirming dermatologist

See ExampleAIAssistedConfirmed.

ExampleAnnotationSeries

Scenario: The system groups the expert annotation and the AI-assisted confirmed annotation into a longitudinal series for the same lesion. Change detection determines that the classification remained stable.

Key elements:

• hasMember references both ExampleMelanomaAnnotation and ExampleAIAssistedConfirmed
• component[changeDetected] is false, indicating no change in classification
• component[changeType] is unchanged
• component[earliestDate] and component[latestDate] define the temporal range

See ExampleAnnotationSeries.

Supporting resources

The examples reference shared Patient, Practitioner, Device, and DocumentReference resources:

• example-patient — Jane Smith, the synthetic patient
• example-dermatologist — Dr. Robert Johnson, the annotating clinician
• example-derm-ai-model — DermAI Classifier v2.1, the AI device
• example-clinical-photo — a DocumentReference for the source clinical photograph (JPEG image of a right upper arm lesion)