Ontology Mapping

A foundational technique that compares the string similarity of concept names, synonyms, and labels to identify potential mappings. This approach leverages:

• Edit distance algorithms like Levenshtein and Jaro-Winkler to quantify character-level differences
• Tokenization and normalization to handle punctuation, case, and word order variations
• Synonym expansion using source terminology thesauri to increase recall

Lexical matching is computationally efficient but struggles with homonyms and concepts that are semantically equivalent yet lexically dissimilar, such as 'hypertension' and 'high blood pressure'.

An advanced alignment technique that uses the formal semantics, hierarchical context, and logical axioms of concepts to determine their degree of similarity. Key components include:

• Graph-based path analysis measuring the distance between concepts within their respective ontological trees
• Property-based comparison evaluating shared attributes, restrictions, and relationships
• Logical axiom analysis using description logic to verify consistency of proposed alignments

This method excels at identifying mappings where lexical overlap is minimal but conceptual equivalence is high, such as aligning 'myocardial infarction' with 'heart attack' through shared parent concepts.

A modern ontology matching technique that uses contextual embeddings from Bidirectional Encoder Representations from Transformers models to capture semantic nuances between concept labels. The process involves:

• Encoding concept names and their definitions into dense vector representations
• Computing cosine similarity between embedding pairs to score potential alignments
• Leveraging pre-trained biomedical models like BioBERT or PubMedBERT for domain-specific accuracy

This approach captures contextual meaning that string-based methods miss, such as understanding that 'renal' and 'kidney' are synonymous in clinical contexts.

A type of ontology alignment that asserts a relationship of logical equality or interchangeability between a concept in a source code system and a concept in a target code system. Equivalence can be:

• Exact: The concepts are fully synonymous and interchangeable in all contexts
• Narrower-to-Broader: The source concept is more specific than the target
• Broader-to-Narrower: The source concept is more general than the target

In FHIR, these relationships are formalized in the ConceptMap resource using equivalence codes such as 'equivalent', 'wider', 'narrower', and 'inexact'.

The hierarchical relationship where one concept is more general than another, such that the broader concept fully encompasses the meaning of the narrower one. This is critical for:

• Query expansion: Retrieving all subclasses of a disease category when searching patient records
• Validation: Ensuring coded data conforms to expected hierarchical constraints
• Automated classification: Using a reasoner to infer new subsumption relationships from asserted axioms

For example, 'Type 2 Diabetes Mellitus' is subsumed by 'Diabetes Mellitus', which is subsumed by 'Endocrine System Disorder' in SNOMED CT.

A workflow where a domain expert reviews, accepts, or rejects algorithmically generated ontology mappings to ensure final accuracy and clinical safety. This process typically involves:

• Confidence score thresholding to triage high-probability matches for automated approval
• Review interfaces displaying source and target concepts side-by-side with hierarchical context
• Mapping provenance capture to record reviewer identity, timestamp, and justification

This hybrid approach combines the scalability of automated matching with the clinical judgment necessary to resolve ambiguous or safety-critical alignments.