Reproduce “GraphMed-LT: Patient-Specific Graph Memory with Latent Clinical Thought Refinement for Multi-Turn Medical Conversations” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhaohan Meng, Zaiqiao Meng, Siwei Liu, Hao Xu, Ke Yuan, Iadh Ounis

GraphMed-LT supports multi-turn medical question answering by building a patient-specific graph from facts gathered across dialogue turns and refining paired evidence and hidden-state thought tokens before the doctor asks again or answers. On iMedQA, the Qwen-2.5-72B doctor achieves 73.0% diagnostic accuracy, versus 67.5% for ProMed. These candidate-answer benchmark results do not establish clinical effectiveness with real patients.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.