What is MedRAG
MedRAG is a medical diagnostic model proposed by a research team at Nanyang Technological University, which enhances the diagnostic capabilities of large language models (LLMs) by combining knowledge graph reasoning. The model constructs a four-layer fine-grained diagnostic knowledge graph that can precisely classify different disease manifestations, and fills gaps in patient information through an active follow-up questioning mechanism. MedRAG improves diagnostic accuracy by 11.32% on real clinical datasets, has good generalization ability, and can be applied to different LLM base models. MedRAG supports multimodal input and can parse symptoms in real time and generate precise diagnostic recommendations.
Main Features of MedRAG
- Precise Diagnostic Support: MedRAG constructs a four-layer fine-grained diagnostic knowledge graph, enabling precise diagnosis based on the key differences among disease manifestations. Through the diagnostic difference knowledge graph search module, the patient's symptoms are matched with diagnostic features in the knowledge graph, precisely locating the most similar symptom nodes and identifying the manifestations most important for disease differentiation, thereby providing strong support for precise diagnosis and personalized treatment plans.
- Intelligent Supplementary Questioning: MedRAG has an active diagnostic questioning mechanism that can automatically generate efficient and precise supplementary questions, helping doctors quickly fill information gaps and improve the accuracy and reliability of diagnosis. When the information provided by the patient is insufficient to distinguish certain diseases, the model is prompted to generate targeted follow-up questions to complete the symptom description.
- Efficient Patient Information Parsing: In the UI interaction design, MedRAG supports multimodal input, including non-intrusive consultation voice monitoring, text input, and electronic health record upload, ensuring that doctors can quickly enter patient information. The system parses symptoms in real time, retrieves similar cases from the local case library, and generates precise diagnostic recommendations through knowledge graph reasoning.
Technical Principles of MedRAG
- Construction of a Four-Layer Fine-Grained Diagnostic Knowledge Graph: This addresses the problem that existing medical knowledge bases are not fine-grained enough and lack comparative information on symptoms of specific diseases. The researchers used techniques such as disease clustering, hierarchical aggregation, semantic embedding, medical expert knowledge, and large language model enhancement to construct a four-layer knowledge graph containing disease categories, subcategories, specific disease names, and disease features.
- Diagnostic Difference Knowledge Graph Search: Used to match the patient's symptoms with diagnostic features in the knowledge graph. The core process includes clinical feature decomposition, breaking down the patient's description into independent symptom representations; clinical symptom matching, calculating the similarity between the patient's symptoms and knowledge graph features to locate the most similar symptom nodes; upward traversal, finding the most relevant disease categories in the knowledge graph; and extraction of key diagnostic features, identifying the representations most important for disease differentiation.
- Knowledge Graph-Guided LLM Reasoning: Traditional RAG relies only on retrieved cases for diagnostic generation, whereas MedRAG enhances the LLM's reasoning ability through a knowledge graph. First, it builds an efficient index based on FAISS to precisely locate clinically similar cases; then it matches key diagnostic features to extract the patient's most discriminative representation information; next, through the LLM, it integrates the retrieved case information, the diagnostic difference knowledge graph, and patient information for joint reasoning to generate precise diagnostic recommendations.
MedRAG Project Address
- Github Repository：https://github.com/SNOWTEAM2023/MedRAG
- arXiv Technical Paper：https://arxiv.org/pdf/2502.04413
Application Scenarios of MedRAG
- Intelligent Health Assistant: MedRAG generates intelligent supplementary questions based on knowledge graph reasoning, helping doctors quickly obtain key diagnostic information.
- Emergency Medicine: In emergency medicine, MedRAG can quickly analyze a patient's symptoms and provide preliminary diagnostic suggestions, shortening the time for diagnosis and treatment.
- Chronic disease management: MedRAG can formulate personalized health management plans based on a patient's medical history and lifestyle habits.
- Medical research: Medical researchers can use MedRAG to quickly obtain the latest medical research materials.
- Medical education: Educational institutions can use MedRAG as a teaching tool to help students better understand and master medical knowledge.
- Hospital online consultation services: Hospitals can use the MedRAG model to provide patients with 24/7 online medical consultation services.