MedReason - A medical reasoning framework developed by institutions including the University of California and Nanyang Technological University.
MedReason is a medical reasoning framework developed by institutions such as the University of California, Santa Cruz, the University of British Columbia, and Nanyang Technological University in Singapore. It leverages knowledge graphs to enhance the application of large language models (LLMs) in medical...
What is MedReason?
MedReason, developed by institutions such as the University of California, Santa Cruz, the University of British Columbia, and Nanyang Technological University, is a medical reasoning framework that enhances the reasoning capabilities of large language models (LLMs) in the medical field based on knowledge graphs. The best model, MedReason-8B, achieves state-of-the-art performance. MedReason transforms clinical question-answer pairs into logical reasoning chains ("thinking paths"), ensuring that each step of reasoning is supported by reliable medical knowledge. The MedReason dataset contains 32,682 question-answer pairs, each with detailed step-by-step explanations. Experiments show that models fine-tuned with MedReason exhibit significant performance improvements across multiple medical benchmarks, particularly in complex clinical scenarios. Expert evaluations have confirmed the accuracy and coherence of the reasoning, providing crucial support for the practical application of medical AI.
MedReason's main functions
- Generate high-quality medical inference dataTransform clinical question-and-answer pairs into logical reasoning chains ("thinking paths") to ensure that each step of reasoning is supported by reliable medical knowledge.
- Improve model performanceBased on supervised fine-tuning (SFT), the performance of LLMs in medical question answering and reasoning tasks is significantly improved, especially in complex clinical scenarios.
- Ensure medical accuracyBased on expert verification and quality filtering mechanisms, the generated reasoning paths are ensured to be medically accurate and coherent.
- Supports multiple medical tasksIt is suitable for a variety of medical question-and-answer and reasoning tasks, including diagnosis, treatment planning, and medical knowledge verification.
MedReason's technical principles
- Medical Entity Extraction and MappingThis method extracts medical entities from questions and answers using a large language model (LLM). Entities are mapped to nodes in a knowledge graph using exact matching, similarity matching, or LLM-based selection.
- Path Search and PruningSearch the knowledge graph for the shortest path connecting the question and answer entities, ensuring the simplicity and logic of the reasoning path. Use LLM to prune paths irrelevant to the current question, retaining only the most relevant reasoning paths.
- Chained reasoning (CoT) generationThe filtered reasoning paths serve as a structural scaffold, guiding LLM to generate chain-of-trace (CoT) explanations based on medical facts. Each step of the reasoning is consistent with the medical knowledge in the knowledge graph, ensuring the accuracy and interpretability of the reasoning.
- Quality FiltrationThe validation process involves having the LLM answer questions using only the generated CoTs and comparing the results to the original answers. CoT samples that fail to produce correct answers are systematically removed to ensure high-quality data.
- Supervisory fine-tuning (SFT)Supervised fine-tuning of LLMs using generated high-quality CoT data improves model performance in medical reasoning tasks.
MedReason's project address
- GitHub repository:https://github.com/UCSC-VLAA/MedReason
- HuggingFace model library:https://huggingface.co/collections/UCSC-VLAA/medreason
- arXiv technical paper:https://arxiv.org/pdf/2504.00993
Application scenarios of MedReason
- Medical Question and Answer SystemDevelop an intelligent medical question-and-answer system to help doctors, medical students, and patients quickly obtain accurate medical information.
- Auxiliary diagnostic toolsIn clinical practice, it serves as an auxiliary diagnostic tool, helping doctors analyze patients' symptoms and medical history to generate possible diagnostic suggestions.
- Medical Education and TrainingUsed in medical education and training to help medical students and practitioners learn complex medical reasoning processes through real-world cases.
- Medical research and knowledge discoveryMedReason is used in medical research to help researchers explore new medical knowledge and treatments.