Relationship structures and semantic type assignments of the UMLS Enriched Semantic Network

Area of Science:

• Medical Informatics
• Computational Linguistics
• Knowledge Representation

Background:

• The Unified Medical Language System (UMLS) Semantic Network (SN) provides a foundational structure for medical terminology.
• Existing limitations in the SN include single inheritance and redundant categorizations, hindering comprehensive semantic abstraction.

Purpose of the Study:

• To introduce the Enriched Semantic Network (ESN) as an extension of the UMLS Semantic Network (SN).
• To present techniques for deriving ESN's relationship structures and semantic type assignments from the SN.
• To address limitations of multiple subsumption and multiple inheritance in semantic network design.

Main Methods:

• Deriving ESN relationship structures by identifying and validating newly inherited relationships, using a blocking mechanism for invalid ones.
• Mapping SN semantic type assignments to the ESN, preserving non-redundant categorizations and preventing new redundancies.
• Automating the derivation of semantic type assignments from the SN to the ESN.

Main Results:

• The ESN incorporates 326 valid newly inherited relationships out of 426.
• Sixteen semantic types exhibit different relationship structures in the ESN compared to the SN.
• The ESN avoids 26,950 redundant categorizations and contains 138 semantic types and 7,303 relationships.

Conclusions:

• The ESN's multiple inheritance enhances relationship structures beyond the SN's capabilities.
• ESN semantic type assignments effectively eliminate existing and potential redundant categorizations.
• The ESN provides a more accurate and comprehensive semantic abstraction of the UMLS Metathesaurus than the SN.