Identifying Mentions of Diseases/Disorders/Medical Issue in Clinical Text:

Dictionary lookup techniques are not always enough to identify the mention of a disease, disorder, or medical issue in clinical text. For example, the phrase “dilatation of the left mitral valve” indicates the existence of a specific medical issue. A dictionary lookup technique alone may only identify the problem; “dilatation” and not the part of anatomy that is experiencing the problem.

Predicting Changes in Systolic Blood Pressure Using Longitudinal Patient Records:

Using a database that contains records of medical events and their dates, along with features from the clinical note associated with each database entry, a logistic regression model is used to predict roughly what the patient’s blood pressure will be at a given future date. This project relates to the more general problem of predicting the occurrence of medical events based on time series data.

Identifying Patients at Risk for Dementia using Semantic and Lexical Features:

Patients that suffer from dementia have subtly different speech patterns than those that do not. Thus there is a system being worked on that could be used to identify if a person is likely to have dementia based on these differences.

Meet the Researcher: Wes Solomon

"I began researching information extraction and event prediction from clinical text and databases based on the suggestion of my advisor, Dr. Rodney Nielsen. This suggestion was based on related work I had done in industry with a company called Gaffey (formerly known as HealthTech) which involved tying status codes to health insurance claims based on natural language data gathered by a web crawler. My specific research topics in this area have been predicting changes in systolic blood pressure using longitudinal patient records, identification of disease/disorder mentions in clinical text, and identifying patients at risk for dementia using semantic and lexical features. Mine and Dr. Nielsen’s work in predicting changes in systolic blood pressure was presented at the i2b2 2014 conference on November 14, 2014.

"In addition to the work I am doing in the clinical domain, I am studying models that estimate the semantic plausibility of syntactically correct statements without explicit definitions or a pre-defined semantic taxonomy. For example, “The building is sad.” is a syntactically correct sentence, but is considered semantically implausible because buildings don’t have emotions. My hope is to expand upon the work previously done in this area by Katrin Erk, Ulrike Padó and Sebastian Padó."