Natural Language Processing in the Clinical Domain: Practical Experiences at the Heidelberg Cardiology Department

Video here

Until today a vast amount of clinical routine data are still stored in unstructured text formats. Natural language processing (NLP) offers methods to convert these data into a machine readable and structured format to make it available for clinical routine and research. At the cardiology department in the University Hospital Heidelberg, we develop methods for automatic entity recognition and text summarization for de-identification and clinical information extraction.

While many state-of-the-art NLP methods are based on supervised deep learning methods, there is a lack of large high-quality annotated clinical text corpora in German. Recent developments in deep learning showed promising results on small sized training data. In this talk I want to present our results of two clinical NLP tasks, automatic de-identification and cardiovascular concept extraction. I want to discuss best-practices in data preparation and evaluate the applicability of deep learning methods for current clinical NLP tasks.

Phillip Richter-Pechanski is a researcher and domain expert in natural language processing and clinical data science. He is a team member of the group of Christoph Dieterich at the Klaus Tschira Institute for Integrative Computational Cardiology and Cardiology Department at the University Hospital Heidelberg. After graduating as M.A. in linguistics at the Free University of Berlin he received his B.A. in computational linguistics with a focus on deep learning and clinical NLP at the Heidelberg University. 

phillip.richter-pechanski at med.uni-heidelberg.de