IlluminateCardio aims to investigate the interpretability of large language models (LLMs) in clinical settings and specifically the cardiovascular domain. Given the steadily growing impact of LLMs in everyday tasks of clinicians such as decision support through simplified access to case files and patient care through dialogue and simplification of medical terminology, the project will develop explanation methods bridging plausible free-text rationales and faithful component analyses and interpretable solutions for information extraction from clinical data sources. Finally, IlluminateCardio will explore the integration of LLMs into digital health interfaces, the evaluation of explanations for different stakeholders, and alignment techniques for more faithful self-explanations of LLMs.

Research Group Lead / Charité

Fellow

Postdoctoral Researcher