Healthcare is a document intensive industry. Critical information lies in a multitude of patient records, clinical guidelines, clinical trials, medical research and publications. In our research in Europe we focus on primarily on identifying medical concepts and facts, and identifying attributes and connections among these facts.
Concept identification allows for semantic-based normalization of extracted information, using standard medical ontologies/terminologies, and helps deal with the heterogeneity of medical texts across document types (e.g. patient records versus clinical guidelines) and across document sources (healthcare providers, academic medical departments, government health organizations, etc.).
Detecting facts along with their attributes and connections, in turn, allows for better generalization in information retrieval, information extraction and data mining. It enables the production of structured information relevant to various use cases and applications, e.g.:
- Risk assessment and early detection of risks that can impact patient safety, e.g. hospital-acquired infections (HAIs). Watch the video.
- Semi-automatic filling of hospital registries and clinical databases for clinical research or administrative purposes
- Chronology extraction from patient records to help improve clinical care process. See the Eureca project.
- Mining medical research publications to detect paradigm shifts and help in the update of clinical guidelines