A SEMANTIC WEB FRAMEWORK FOR REPRESENTING, LINKING AND ANALYZING MEDICAL DATA FOR OPTIMIZING LABORATORY UTILIZATION

Abstract

In this thesis, we investigate semantic web based methods for representing, linking and analyzing medical data. The main challenge addressed in this work is the transformation of data stored in a relational database to an ontological model that allows to represent as RDF triples and to link the data with external data sources using linked data principles. We have implemented a semantic analytics framework that comprises the following elements: (a) Domain-specific ontology to represent the data model and data inference. (b) RDMS data extraction using a domain-specific ontology (TBOX) based on the relational database schema; (c) Ontology instantiation (ABOX) that involves converting the relational data in terms of RDF triples. A key feature of our approach is the data is not physically migrated from the RDBS to RDF, rather we dynamically materialize the RDF triples thus avoiding the creation of a large RDF dump; (d) Linking the RDF data with available open data in RDF format using ontology-based concept alignments; and (e) Semantic analytics using SPARQL to identify semantically-salient patterns within the data. We have applied our semantic analytics data to analyze pathology lab data (over 5 years of pathology order data), where we were able to identify prevalent order-sets inherent within the data, and we also evaluated the change in the frequent order-sets over a five year time period.