Abstract
Healthcare professionals spend a significant portion of their time interacting with electronic health records (EHRs), a task that has become increasingly burdensome due to the exponential growth of healthcare data. This is certainly true in the data rich setting of the intensive care unit (ICU). This project addresses the urgent need to reduce the administrative burden on care providers by developing a (non-English) multimodal question-answering (Q&A) system with a proof of concept for Dutch-speaking clinical practices. Leveraging advancements in large language models (LLMs), Retrieval-Augmented Generation (RAG), and knowledge graphs, the system aims to enhance EHR usability and support evidence-based decision-making. The research has four objectives: (1) integrate clinical patient notes into RAG systems while addressing text anonymization challenges, (2) enhance clinical question answering by combining English and non-English medical literature, (3) develop retrieval methods to extract and summarize relevant clinical time series data, and (4) build and validate a clinical Q&A system that integrates multiple data sources, refining it through real-world feedback from care providers. By dynamically combining text, biomedical knowledge, and time series data, the system will provide a holistic view of the patient while streamlining information retrieval and presentation.
Researcher(s)
Research team(s)
Project type(s)