Dr. Dmitriy Dligach, a newly hired assistant professor at the computer science department at the Loyola Lakeshore Campus, will present a seminar titled "Semantic Analysis of Clinical Texts” on March 8th for our department at the conference room. The abstract is also attached. Before he joined Loyola in January this year, he was a researcher at Boston Children’s Hospital and Harvard Medical school doing work in electronic medical record data mining.
His abstract is as follows:
It is often estimated that 80% of clinical data today is stored in an unstructured form, mostly as electronic health records. Within this corpus of text lies a vast amount of valuable information that can be leveraged for phenotyping, pharmacogenomic studies, clinical studies, and clinical decision support, ultimately improving patient care and reducing healthcare costs. Until recently, this wealth of information could not be analyzed but with the advent of Natural Language Processing (NLP) it became possible to turn this data into a structured form which can be subsequently used for data mining. In this talk, I will discuss various approaches to semantic analysis of clinical narratives. I will begin by describing a coarse semantic analysis task known as phenotyping. The best-performing approaches to phenotyping currently heavily rely on manually annotated data. I will report on my experiments with active learning, a technique that has the potential to drastically reduce the reliance on manually annotated data. I will then go a level deeper and describe a more fine-grained approach to semantic analysis of electronic health records, which involves discovering UMLS relations between clinical entities. I will conclude with an outline of another relation extraction task which involves extracting temporal relations between clinical events and report on the availability of open-source software for deriving clinical timelines from electronic health records.