Researchers have developed REFINE, a novel framework designed to create personalized medical concept representations from electronic health records (EHRs). This approach refines text-attributed knowledge graphs (TKGs) by constructing patient-specific temporal graphs and using a reinforcement learning policy to determine the optimal KG context for each observed code. A heterogeneous graph neural network then captures structural dependencies, while a frozen large language model semantically refines these representations using graph-aware soft prompts. Experiments conducted on the MIMIC-III and MIMIC-IV datasets demonstrate that REFINE significantly enhances various EHR backbones and outperforms existing baseline methods. AI
IMPACT This research could lead to more accurate patient diagnoses and treatment plans by improving how AI models understand and utilize medical data.
RANK_REASON The cluster contains an academic paper detailing a new method for medical concept representation. [lever_c_demoted from research: ic=1 ai=1.0]
- electronic health records
- graph neural network
- MIMIC-III
- MIMIC-IV
- Mohsen Nayebi Kerdabadi
- REFINE
- temporal knowledge graph
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
