ELMTEX: Fine-Tuning LLMs for Structured Clinical Information Extraction. A Case Study on Clinical Reports

Europe’s healthcare must improve interoperability and embrace solutions to unlock the value of legacy clinical data. We used LLMs to transform unstructured clinical reports into structured records. We built a complete workflow, including a UI, and benchmarked various LLM sizes through both prompt engineering and fine-tuning. Our fine-tuned smaller models matched or even surpassed the larger ones, making them ideal for settings with limited computational resources. Finally, we validated a novel dataset of annotated English and German translations of clinical summaries using automated metrics alongside expert manual review.