Prof. Stefan Schulz (Medical University of Graz, Austria): Abstract
Clinical narratives play a pivotal role in healthcare. To unlock their full potential, information must be extracted using NLP and brought into a structure that can be queried. However, clinical language tends to compact and context-dependent, and laden with medical jargon and short forms. These idiosyncrasies are particularly challenging for automated information extraction. The use of language models for clinical information extraction requires clinical corpora annotated by humans. These annotations serve as the foundation for training, fine-tuning and validating AI-based NLP approaches.
However, annotators need guidance to produce reproducible annotations. Therefore, clinical documentation standards like SNOMED CT and FHIR as well as elaborated annotation guidelines are needed. The talk will highlight current clinical annotation activities and sketch the representation of annotation results as knowledge graphs that are compatible with clinical standards and are built on ontological principles.