Can SNOMED CT be squeezed without losing its shape?

Area of Science:

• Ontology engineering
• Biomedical informatics
• Knowledge representation

Background:

• SNOMED CT's large size poses challenges for some biomedical applications.
• Smaller, representative subsets are often preferred for efficiency.
• Extracting balanced modules that maintain SNOMED CT's structural properties is crucial but underexplored.

Purpose of the Study:

• To investigate the feasibility of extracting balanced modules from SNOMED CT.
• To evaluate an iterative algorithm for creating representative ontology subsets.
• To assess the preservation of sub-hierarchy shape in extracted modules.

Main Methods:

• Employed a graph-traversal modularization approach using an input signature.
• Implemented an iterative algorithm that dynamically adjusted the signature.
• Defined module balance by measuring sub-hierarchy error and achieving convergence (residual sum of squares <1).

Main Results:

• The algorithm converged in seven iterations, starting with a 2000-concept signature.
• Successfully extracted a module representing 4.7% of the original SNOMED CT size.
• Seven sub-hierarchies showed minor over/under-representation (1-8%).

Conclusions:

• Balanced modules can be extracted from large terminologies like SNOMED CT using iterative graph-traversal methods.
• Dynamic signature adjustment and tolerance for minor hierarchy variations are key.
• SNOMED CT can be reduced to <5% of its size with <8% shape deviation, suitable for many use cases.