AI4MedCode: AI-Supported ICD Coding for German Hospitals

Project overview

AI4MedCode delivered a usable NLP-based solution for German hospitals, assisting staff with ICD coding and billing. The project was a collaboration between the University of Stuttgart and Klinikum Stuttgart. I contributed in Research and Development at QUIBIQ GmbH until October 2024.

My contributions

• Developed NLP tools to support ICD coding by hospital staff.
• Used generative AI to create and annotate synthetic German clinical narratives for biomedical Named Entity Recognition (NER), addressing the shortage of annotated data in low-resource settings.
• Implemented a Wikidata-based biomedical knowledge base for entity linking in German.

Related publications

Leveraging Wikidata for Biomedical Entity Linking in a Low-Resource Setting: A Case Study for German

This paper describes a German biomedical entity-linking approach built on a Wikidata knowledge base using UMLS information. It compares multiple linking and retrieval methods and finds that strong language-specific knowledge-base coverage improves performance.

The Aluminum Standard: Using Generative Artificial Intelligence Tools to Synthesize and Annotate Non-Structured Patient Data

This study explores generating synthetic German clinical narratives from disease and comorbidity patterns, then annotating them to train and evaluate biomedical NER models. This offers a way to investigate model training when real annotated clinical text is scarce.

Both publications are also listed on the Publications page, with links to the papers and related research resources.

For additional project information, see the AI4MedCode project page at the University of Stuttgart.