Semantic concept extraction from electronic medical records for enhancing information retrieval performance

With the healthcare industry increasingly using EMRs, there emerges an opportunity for knowledge discovery within the healthcare domain that was not possible with paper-based medical records. One such opportunity is to discover UMLS concepts from EMRs. However, with opportunities come challenges that need to be addressed. Medical verbiage is very different from common English verbiage and it is reasonable to assume extracting any information from medical text requires different protocols than what is currently used in common English text. This thesis proposes two new semantic matching models: Term-Based Matching and CUI-Based Matching. These two models use specialized biomedical text mining tools that extract medical concepts from EMRs. Extensive experiments to rank the extracted concepts are conducted on the University of Pittsburgh BLULab NLP Repository for the TREC 2011 Medical Records track dataset that consists of 101,711 EMRs that contain concepts in 34 predefined topics. This thesis compares the proposed semantic matching models against the traditional weighting equations and information retrieval tools used in the academic world today.