EpiGraph: Knowledge Graph for Epilepsy Clinical Reasoning
A team of researchers has created EpiGraph, an extensive epilepsy knowledge graph that incorporates 48,166 peer-reviewed articles and seven clinical resources, resulting in a diverse graph featuring 24,324 entities and 32,009 evidence-based triplets organized into five clinical layers. The related benchmark, EpiBench, outlines five clinically relevant tasks: clinical decision-making, EEG report generation, pharmacogenomic precision medicine, treatment recommendations, and deep research planning. Assessments of six LLMs in both standard and Graph-RAG conditions indicate that the integration of EpiGraph consistently enhances performance. This initiative aims to promote evidence-based reasoning in the diagnosis and treatment of epilepsy.
Key facts
- EpiGraph integrates 48,166 peer-reviewed papers and seven clinical resources.
- The graph contains 24,324 entities and 32,009 evidence-grounded triplets.
- Five clinical layers are included in the knowledge graph.
- EpiBench defines five clinically motivated tasks.
- Six LLMs were evaluated under standard and Graph-RAG settings.
- Integrating EpiGraph consistently improves LLM performance.
- The work focuses on epilepsy diagnosis and treatment reasoning.
- The paper is published on arXiv with ID 2605.09505.
Entities
Institutions
- arXiv