⭐ Star this repository if you use Medical RAG, GraphRAG, clinical QA, biomedical retrieval, or multilingual AI.
Tip
The Definitive Knowledge Hub for Medical AI
Welcome to a meticulously curated collection of 250+ research papers, datasets, and benchmarks. This repository bridges the gap between Structured Medical Knowledge (Knowledge Graphs) and Generative Reasoning (LLMs), focusing on:
- Retrieval-Augmented Generation (RAG): Techniques to ground AI in factual clinical data.
- Multilingual Equity: Ensuring medical AI works across diverse languages and cultures.
- Complex Reasoning: Moving from simple Q&A to multi-hop clinical decision support.
Note
How to Navigate
Organized taxonomically to help you find exactly what you need:
- By Method: Looking for GraphRAG, PEFT, or Visual QA?
- By Domain: Interested in Mental Health, Legal, or Finance?
Use the Table of Contents below to jump straight to your area of interest.
- 🏥 Awesome Medical & Multilingual QA/RAG Research Papers
Important
Recommended Reading Path for Medical RAG
Start with the survey and benchmark papers, then move to GraphRAG/KG-enhanced systems, then clinical/EHR RAG, and finally safety-focused evaluation. This makes the list easier to scan for newcomers and more useful for researchers comparing medical RAG pipelines.
| Track | Best Starting Points | Why It Matters |
|---|---|---|
| Medical RAG Foundations | MIRAGE / MedRAG Toolkit, RAG in Biomedicine Survey, MRAG | Gives a practical baseline for corpora, retrievers, prompting, and evaluation. |
| GraphRAG & Knowledge Graphs | Medical Graph RAG, MedRAG, KG-Rank, KRAGEN, DR.KNOWS | Adds structured biomedical relations, path reasoning, and explainable evidence. |
| Clinical & EHR RAG | RGAR, i-MedRAG, ClinicalRAG, MedBioRAG | Connects retrieved literature with patient-specific or clinical-context evidence. |
| Safety & Trustworthiness | RAG², RAG-X, Contradictions in Context, MedRGB | Measures grounding, context use, outdated evidence, and deceptive accuracy. |
| Multilingual & Multimodal RAG | Medical mT5, Japanese KG-RAG, MED-VRAG, EHRXQA | Extends medical QA beyond English text-only retrieval. |
Tip
For more GitHub stars: keep the first screen attractive, add fresh 2025-2026 papers near the top of each section, label code/dataset/project links clearly, and make it obvious why this list is different from generic RAG lists: medical safety, multilingual equity, and knowledge-graph grounding.
Important
Foundational Models & Clinical Adaptations
This section aggregates Large Language Models (LLMs) rigorously adapted for the medical domain. It encompasses a spectrum of methodologies from Continued Pre-training on biomedical corpora to Instruction Fine-tuning and Parameter-Efficient Fine-Tuning (PEFT). These models bridge the gap between general-purpose AI and the high-stakes requirements of clinical reasoning and healthcare NLP.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Medical Adaptation | HuatuoGPT-II, One-stage Training for Medical Adaptation of LLMs |    |
One-stage training for medical adaptation. |
| QA System | SPBERTQA: A Two-Stage Question Answering System Based on Sentence Transformers for Medical Texts |   |
2-Stage QA based on Sentence Transformers. |
| Medical LLM | Medical mT5: An Open-Source Multilingual Text-to-Text LLM for The Medical Domain |    |
Multilingual T5 for medical domain. |
| Closed-Domain QA | Large Language Models Encode Clinical Knowledge (Med-PaLM) |   |
Seminal Med-PaLM paper. |
| Medical Tuning | ClinicalGPT: Large Language Models Finetuned with Diverse Medical Data and Comprehensive Evaluation |   |
Fine-tuning on diverse data. |
| Clinical Fine-Tune | LlamaCare: An Instruction Fine-Tuned Large Language Model for Clinical NLP |  |
Instruction fine-tuned LLaMA. |
| Virtual Assistant | KG-Infused LLM for Virtual Health Assistant: Accelerated Inference and Enhanced Performance |  |
KG-infused LLM assistant. |
| Safe MedLLM | Medical Graph RAG: Towards Safe Medical Large Language Model via Graph RAG |   |
MedGraphRAG framework. |
| Clinical PEFT | Parameter-Efficient Fine-Tuning of LLaMA for the Clinical Domain |   |
Clinical LLaMA-LoRA. |
| French Model | DrBERT: A Robust Pre-trained Model in French for Biomedical and Clinical Domains |  |
BERT for French Bio/Clinical. |
| Spanish Biomed | Pretrained Biomedical Language Models for Clinical NLP in Spanish |  |
Beteo / RoBERTa-es. |
| Vietnamese Health | ViHealthBERT: Pre-trained Language Models for Vietnamese in Health Text Mining |  |
ViHealthBERT. |
| TCM Tuning | HuaTuo: Tuning LLaMA Model with Chinese Medical Knowledge |  |
HuaTuo (LLaMA tuning). |
| Complex Reasoning | HuatuoGPT-o1: Towards Medical Complex Reasoning with LLMs |  |
O1-like medical reasoning. |
| Japanese 70B | 70B-parameter Large Language Models in Japanese Medical Question-Answering | |
70B Japanese Medical LLM. |
| Clinical Mamba | ClinicalMamba: A Generative Clinical Language Model on Longitudinal Clinical Notes |  |
Mamba model for clinical notes. |
| Trilingual LLM | ELAINE-medLLM: Lightweight English Japanese Chinese Trilingual Large Language Model |  |
En-Jp-Zh Medical LLM. |
| Note Gen | A Continued Pretrained LLM Approach for Automatic Medical Note Generation |   |
Auto note generation. |
| Clinical Needs | Do We Still Need Clinical LLMs? |  |
Questioning need for specialized LLMs. |
| French Biomed | AliBERT: A Pre-trained Language Model for French Biomedical Text | |
AliBERT. |
| Portuguese | BioBERTpt: A Portuguese Neural Language Model for Clinical NER | |
BioBERTpt. |
| Portuguese Card | CardioBERTpt: Transformer-based Models for Cardiology Language Representation | |
CardioBERTpt. |
| Spanish Clinical | Clinical Flair: A Pre-Trained Language Model for Spanish Clinical NLP | |
Clinical Flair. |
| Spanish Pre-train | Development of pre-trained language models for clinical NLP in Spanish |  |
Pre-trained models for Spanish. |
| Vietnamese Biomed | ViPubmedDeBERTa: A Pre-trained Model for Vietnamese Biomedical Text |  |
ViPubmedDeBERTa. |
| Chinese LLM | Zhongjing: Enhancing Chinese Medical LLM through Expert Feedback |   |
Zhongjing model. |
| Textbook Aug | Augmenting Black-box LLMs with Medical Textbooks for Biomedical QA |   |
Textbook augmentation. |
| Collaborative | MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning |  |
Multi-agent collaboration. |
| BioBERT | MSQ-BioBERT: Ambiguity Resolution to Enhance BioBERT Medical QA |  |
Ambiguity resolution. |
| EHR Repr | MHGRL: An Effective Representation Learning Model for Electronic Health Records | |
Representation learning. |
| CamemBERT | CamemBERT-bio: Leveraging Continual Pre-training on French Biomedical Data |  |
French domain adaptation. |
| Japanese Eval | Evaluating GPT-4 and ChatGPT on Japanese Medical Licensing Examinations |  |
IgakuQA Benchmark. |
| Fine-tuning | Effects of Information Masking in the Task-Specific Finetuning of a Transformers-Based Clinical Question-Answering Framework | |
Info masking in fine-tuning. |
| Embeddings | MedITA_embeddings: Combining Contrastive Learning and Knowledge Graph Embeddings to develop medical word embeddings |  |
KG + Contrastive embeddings. |
| Medical RL | Breaking Reward Collapse: Adaptive Reinforcement for Open-ended Medical Reasoning with Enhanced Semantic Discrimination |  |
Adaptive RL for medical reasoning. |
| Reasoning RL | CAPO: Reinforcing Consistent Reasoning in Medical Decision-Making |  |
Consistent reasoning via RL. |
| Medical RL | Fleming-R1: Toward Expert-Level Medical Reasoning via Reinforcement Learning |  |
Expert-level reasoning via RL. |
| Unified Reasoning | Med-U1: Incentivizing Unified Medical Reasoning in LLMs via Large-scale Reinforcement Learning |  |
Unified reasoning via RL. |
| Reasoning Emergence | Med-RLVR: Emerging Medical Reasoning from a 3B Base Model via Reinforcement Learning |  |
RL on 3B model. |
| Diagnosis Reasoning | MedReason-R1: Learning to Reason for CT Diagnosis with Reinforcement Learning and Local Zoom |  |
RL for CT diagnosis. |
| Patient-Doctor | Dr. Copilot: A Multi-Agent Prompt Optimized Assistant for Improving Patient-Doctor Communication in Romanian |  |
Romanian multi-agent assistant. |
| Medical Advice | PIE-Med: Predicting, Interpreting and Explaining Medical Recommendations | |
Predict-Interpret-Explain. |
| Error Detection | Maven at MEDIQA-CORR 2024: Leveraging RAG and Medical LLM for Error Detection and Correction in Medical Notes | |
Error detection in notes. |
| Structured Ext | Structured Extraction of Real World Medical Knowledge using LLMs for Summarization and Search |  |
Structured extraction. |
| Machine Reading | Towards Medical Machine Reading Comprehension with Structural Knowledge and Plain Text | |
KMQA model. |
| Virtual Patient | A Virtual Patient Dialogue System Based on Question-Answering on Clinical Records | |
Virtual Patient Dialogue. |
| Herbal Medicine | ViHerbQA: A Robust QA Model for Vietnamese Traditional Herbal Medicine | |
Vietnamese Herbal Medicine QA. |
| Rule-Free | A Rule-Free Approach for Cardiological Registry Filling from Italian Clinical Notes with Question Answering Transformers | |
Cardiological registry filling. |
| Clinical Cases | Using Structured Health Information for Controlled Generation of Clinical Cases in French | |
Generating French clinical cases. |
| Zero-Shot Clinical | Zero-Shot Clinical Questionnaire Filling From Human-Machine Interactions |  |
Zero-shot questionnaire filling. |
| Safety Reports | Automatic Job Safety Report Generation using RAG-based LLMs | |
Job safety report generation. |
| Biomed Reasoning | Hierarchical Representation-based Dynamic Reasoning Network for Biomedical Question Answering (HDRN) |  |
HDRN for Biomedical QA. |
| QA Consistency | Are Large Language Models Consistent in Health-Related Question Answering? |  |
Consistency check in Health QA. |
| Entity Masking | Boosting Low-Resource Biomedical QA via Entity-Aware Masking Strategies | |
Entity-aware masking strategies. |
| Biomed QA | Biomedical Question Answering Systems into Practice: A Survey |   |
Survey of BioMed QA systems. |
| Search / Dialogue | Anticipating Follow-Up Questions in Exploratory Information Search |  |
Follow-up question prediction. |
| Exam Generation | From Human Days to Machine Seconds: Automatically Answering and Generating Machine Learning Final Exams |  |
ML Exam generation/answering. |
| French NER | A Benchmark Evaluation of Clinical Named Entity Recognition in French |  |
Clinical NER in French. |
| Chinese Medical | Benchmarking Large Language Models on CMExam: A Comprehensive Chinese Medical Examination Dataset |  |
CMExam benchmark. |
| DrQA | Reading Wikipedia to Answer Open-Domain Questions (DrQA) |  |
Seminal DrQA paper. |
| Few-shot QA | Gotta: Generative Few-shot Question Answering by Prompt-based Cloze Data Augmentation |  |
Prompt-based data augmentation. |
| Spoken QA | A Framework for Automatic Generation of Spoken Question-Answering Data | |
Spoken QA data generation. |
| Non-factoid QA | MICRON: Multigranular Interaction for Contextualizing RepresentatiON in Non-factoid Question Answering |  |
MICRON model. |
| Synthetic QA | Training Generative Question-Answering on Synthetic Data Obtained from an Instruct-tuned Model | |
Synthetic training data. |
| Hebrew NLP | Transformer-based Hebrew NLP models for Short Answer Scoring in Biology |  |
Hebrew Short Answer Scoring. |
| Data Selection | Data Selection for Language Models: A Survey |  |
Survey on data selection. |
| QA Eval | Evaluating Open-QA Evaluation |  |
Evaluation of QA evaluation. |
| Explainable QA | Explainable Bilingual Medical-Question-Answering Model Using Ensemble Learning Technique |  |
Bilingual Explainable QA. |
| Hint Gen | HintQA: Exploring Hint Generation Approaches in Open-Domain Question Answering |  |
Hint generation for ODQA. |
| Memory QA | Omne-R1: Learning to Reason with Memory for Multi-hop Question Answering |  |
Reasoning with Memory. |
Important
Breaking Language Barriers
This section is dedicated to democratizing AI access through Cross-Lingual Transfer and Multilingual RAG. It features benchmarks for under-represented languages (e.g., Amharic, Tigrinya, Kazakh), techniques for cultural alignment, and strategies to bridge the performance gap between high-resource and low-resource linguistic domains.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Multilingual Adapters | Adapters for Enhanced Modeling of Multilingual Knowledge and Text |  |
Enhanced modeling adapters. |
| Entity Alignment | Aligning Cross-Lingual Entities with Multi-Aspect Information |  |
Multi-aspect alignment. |
| Arabic Medical | AraMed: Arabic Medical Question Answering using Pretrained Transformer Language Models |  |
Arabic Medical QA. |
| Multi-Domain QA | M2QA: Multi-domain Multilingual Question Answering |  |
M2QA benchmark. |
| Cross-Lingual | XLM-K: Improving Cross-Lingual Language Model Pre-training with Multilingual Knowledge |  |
Pre-training with multilingual knowledge. |
| Prompt Transfer | Cross-Lingual Transfer for Natural Language Inference via Multilingual Prompt Translator |  |
Multilingual prompt translation. |
| Thai LLM | Representing the Under-Represented: Cultural and Core Capability Benchmarks for Thai LLMs | |
Thai cultural/core benchmarks. |
| ICL Alignment | Improving In-context Learning of Multilingual Generative Language Models with Cross-lingual Alignment | |
Cross-lingual alignment for ICL. |
| Kazakh QA | KazQAD: Kazakh Open-Domain Question Answering Dataset |  |
Kazakh open-domain QA. |
| Multilingual RAG | Not All Languages are Equal: Insights into Multilingual Retrieval-Augmented Generation |  |
Linguistic inequalities in RAG. |
| Machine Translation | Enhancing Machine Translation Experiences with Multilingual Knowledge Graphs |  |
MT with Multilingual KG. |
| Amharic QA | Low Resource Question Answering: An Amharic Benchmarking Dataset (Amh-QuAD) |  |
Amh-QuAD / AmQA. |
| Tigrinya QA | Question-Answering in a Low-resourced Language: Benchmark Dataset and Models for Tigrinya |  |
Tigrinya QA benchmark. |
| Portuguese QA | Pirá: A Bilingual Portuguese-English Dataset for Question-Answering | |
Pirá dataset. |
| Zero-shot CoT | AutoCAP: Towards Automatic Cross-lingual Alignment Planning for Zero-shot Chain-of-Thought |  |
AutoCAP. |
| Cultural QA | NativQA: Multilingual Culturally-Aligned Natural Query for LLMs |  |
Culturally-aligned questions. |
| Swedish NLP | A Benchmark for Swedish Clinical Natural Language Processing |  |
Swedish Clinical NLP benchmark. |
| Transferability | On the Cross-lingual Transferability of Monolingual Representations |   |
Monolingual transferability. |
| Linguistic Inc | URIEL+: Enhancing Linguistic Inclusion and Usability in a Typological and Multilingual Knowledge Base |  |
URIEL+ Knowledge Base. |
| Prompt Transfer | Is Prompt Transfer Always Effective? An Empirical Study of Prompt Transfer for Question Answering | |
Empirical study on prompt transfer. |
| Romanian Eval | Evaluation of Language Models on Romanian XQuAD and RoITD datasets |  |
Romanian XQuAD/RoITD. |
Warning
Bridging Structured & Unstructured Knowledge
This section dives into the complex integration of Knowledge Graphs (KGs) with Large Language Models. It covers advanced reasoning tasks such as Multi-hop Reasoning, Subgraph Extraction, and Neuro-Symbolic approaches. Papers here explore how to ground LLM generation in factual graph structures to improve accuracy and interpretability.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| KG Alignment | MRAEA: An Efficient and Robust Entity Alignment Approach for Cross-lingual KG | |
Cross-lingual KG alignment. |
| Reasoning Eval | DIVKNOWQA: Assessing the Reasoning Ability of LLMs via Open-Domain QA over KB and Text |  |
Reasoning over KB and Text. |
| KG + LLM | ChatDoctor: A Medical Chat Model Fine-Tuned on LLaMA Using Medical Domain Knowledge |  |
LLaMA + Medical KG. |
| KG Prompting | KG-CoT: Chain-of-Thought Prompting of LLMs over Knowledge Graphs |  |
CoT over Knowledge Graphs. |
| KG Sync | Domain Knowledge Exploration by Synchronizing Knowledge Graph and LLMs |  |
Synchronizing KG and LLMs. |
| Dynamic Reasoning | Dynamic Hierarchical Reasoning with Language Model and Knowledge Graph (DRLK) |  |
Hierarchical reasoning with KG. |
| Subgraph Reasoning | ReasoningLM: Enabling Structural Subgraph Reasoning in Pre-trained Language Models |  |
Structural subgraph reasoning. |
| Graph Reasoning | RoG: Reasoning on Graphs (Faithful and Interpretable LLM Reasoning) |   |
Reasoning paths for QA. |
| Graph Reasoning | KG-GPT: A General Framework for Reasoning on Knowledge Graphs Using LLMs |  |
KG-GPT framework. |
| KG Integration | Bridging the Gap: Integrating Knowledge Graphs into LLMs for Complex QA | |
KG + LLM for complex QA. |
| Multilingual KG | Joint Completion and Alignment of Multilingual Knowledge Graphs |  |
JMAC model. |
| Knowledge Injection | KIMERA: Injecting Domain Knowledge into Vacant Transformer Heads | |
Retraining vacant heads. |
| Quotes KG | QuoteKG: A Multilingual Knowledge Graph of Quotes |  |
KG of quotes (55 langs). |
| TCM KG | Traditional Chinese Medicine Knowledge Graph Construction Based on LLMs | |
TCM KG construction. |
| Updating KG | Up To Date: Automatic Updating Knowledge Graphs Using LLMs |  |
Auto updating KGs. |
| Semi-supervised | Semi-supervised Entity Alignment via Joint Knowledge Embedding (KECG) | |
KECG model. |
| Alignment Transf. | Multi-Modal Knowledge Graph Transformer Framework for Entity Alignment | |
Meaformer. |
| Contrastive | Multi-modal Contrastive Representation Learning for Entity Alignment | |
MCLEA model. |
| KBQA Framework | ChatKBQA: A Generate-then-Retrieve Framework for KBQA with Fine-tuned LLMs |  |
Generate-then-Retrieve. |
| Few-Shot KBQA | KB-BINDER: A Unified Semantically Parsing Framework for KBQA |  |
Few-shot semantic parsing. |
| Semantic Parsing | LLM-based Semantic Parsing for Conversational QA over Knowledge Graphs |  |
Semantic parsing for ConvQA. |
| Complex QA | KQA Pro: A Dataset with Explicit Compositional Programs for Complex KBQA |  |
KQA Pro dataset. |
| Rule-Guided | Rule-KBQA: Rule-Guided Reasoning for Complex KBQA with LLMs | |
Rule-guided reasoning. |
| Medical KG | Research on Medical Question Answering System Based on Knowledge Graph | |
Neo4j-based Medical KG. |
| Knowledge Injection | Infusing Disease Knowledge into BERT for Health QA and Inference |  |
Infusing knowledge into BERT. |
| Journal KG | Construction of Journal Knowledge Graph Based on Deep Learning and LLM | |
Journal KG construction. |
| Alzheimer's QA | Dynamic Co-Augmentation of Large Language Models and Knowledge Graphs for Question Answering in Alzheimer's Disease Scientific Literature |  |
Co-augmentation for Alzheimer's. |
| Aviation QA | Knowledge Graph – Deep Learning: A Case Study in Question Answering in Aviation Safety Domain | |
Aviation safety KG-QA. |
| Brain Disorders | Knowledge Graph and Its Application in the Study of Neurological and Mental Disorders |   |
KG for brain disorders. |
| Chronic Diseases | Research on Question Answering over Knowledge Graph of Chronic Diseases | |
Chronic disease KG. |
| Entity Alignment | Exploring the Impacts of Feature Fusion Strategy in Multi-modal Entity Alignment | |
Feature fusion in alignment. |
| Object Prediction | Evaluating Prompt-based Question Answering for Object Prediction in the Open Research Knowledge Graph | |
Prompt-based object prediction. |
Tip
Augmenting Generation with External Medical Knowledge
This section focuses on Retrieval-Augmented Generation (RAG) systems that reduce hallucinations, improve clinical grounding, and connect LLMs with biomedical literature, EHRs, guidelines, ontologies, and knowledge graphs. The papers are grouped by use case so readers can quickly compare survey/benchmark work, GraphRAG, clinical and EHR retrieval, safety evaluation, and retrieval optimization.
Warning
Medical RAG is not only retrieval. Strong systems must also handle outdated evidence, contradictory sources, patient-specific context, citation quality, and clinical safety. Prefer papers that report retrieval quality, answer faithfulness, source attribution, robustness, and human/clinical evaluation.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Benchmark / Toolkit | Benchmarking Retrieval-Augmented Generation for Medicine |   |
MIRAGE benchmark and MedRAG toolkit for medical QA. |
| BioMed Survey | Retrieval-Augmented Generation in Biomedicine: A Survey of Technologies, Datasets, and Clinical Applications |  |
Survey of biomedical RAG methods, datasets, and clinical uses. |
| Clinical Review | Retrieval-Augmented Generation in Medicine: A Scoping Review of Technical Implementations, Clinical Applications, and Ethical Considerations |  |
Clinical applications, implementation patterns, and ethics. |
| Healthcare Review | Retrieval augmented generation for large language models in healthcare: A systematic review |  |
Healthcare RAG review across retrieval, augmentation, and generation. |
| Meta-analysis | Retrieval-Augmented Generation and Large Language Models in Biomedicine: A Systematic Review, Meta-analysis, and Clinical Development Guidelines |  |
Evidence synthesis and clinical development guidance. |
| MRAG Benchmark | MRAG: Benchmarking Retrieval-Augmented Generation for Bio-medicine |  |
English/Chinese biomedical RAG benchmark and toolkit. |
| MedRGB | Comprehensive and Practical Evaluation of Retrieval-Augmented Generation Systems for Medical Question Answering |  |
Tests sufficiency, integration, robustness, and noisy retrieval. |
| RAG Design | A Systematic Study of Retrieval Pipeline Design for Retrieval-Augmented Medical Question Answering |  |
Retrieval design choices on MedQA with modest compute. |
| RAG Chatbots | Automated Question-Answer Generation for Evaluating RAG-based Chatbots |  |
Auto QA generation for healthcare RAG evaluation. |
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Safe MedGraphRAG | Medical Graph RAG: Towards Safe Medical Large Language Model via Graph Retrieval-Augmented Generation | |
Triple graph construction and U-Retrieval for safer medical responses. |
| Healthcare Copilot | MedRAG: Enhancing Retrieval-Augmented Generation with Knowledge Graph-Elicited Reasoning for Healthcare Copilot |  |
Diagnostic KG reasoning with EHR retrieval and follow-up questions. |
| Ranking/Re-ranking | KG-Rank: Enhancing LLMs for Medical QA with Knowledge Graphs and Ranking Techniques |   |
KG-based retrieval/ranking for medical QA. |
| Biomedical Problem Solving | KRAGEN: A Knowledge Graph-Enhanced RAG Framework for Biomedical Problem Solving Using Large Language Models |   |
KG + RAG + graph-of-thought prompting. |
| Diagnosis KG | DR.KNOWS: Leveraging Medical Knowledge Graphs Into Large Language Models for Diagnosis Prediction |  |
UMLS paths aligned with patient-specific EHR information. |
| Biomedical Agent | ESCARGOT: An AI Agent Leveraging LLMs, Dynamic Graph of Thoughts, and Biomedical Knowledge Graphs for Enhanced Reasoning | |
Agentic KG reasoning with dynamic graph-of-thoughts. |
| Structured Reflection | Structured Reflective Reasoning for Precise Medical Knowledge Graph Retrieval-Augmented Generation |  |
Type-aware pre-anchoring and reflective KG reasoning. |
| KG-RAG | Knowledge Graph-extended Retrieval Augmented Generation (KG-RAG) |  |
KG-extended RAG with biomedical grounding potential. |
| TCM GraphRAG | OpenTCM: A GraphRAG-Empowered LLM-based System for Traditional Chinese Medicine |  |
GraphRAG for TCM knowledge and consultation. |
| Japanese KG-RAG | Exploring the Role of Knowledge Graph-Based RAG in Japanese Medical Question Answering with Small-Scale LLMs |  |
KG-RAG for Japanese medical QA. |
| Region-First | ReGraM: Region-First Knowledge Graph Reasoning for Medical QA |  |
Region-first graph reasoning for medical QA. |
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Heterogeneous Retrieval | ClinicalRAG: Enhancing Clinical Decision Support through Heterogeneous Knowledge Retrieval |  |
Clinical decision support with heterogeneous evidence. |
| Factual + Conceptual | RGAR: Recurrence Generation-Augmented Retrieval for Factual-aware Medical Question Answering |  |
Iteratively combines EHR factual knowledge with corpus knowledge. |
| Iterative RAG | Improving Retrieval-Augmented Generation in Medicine with Iterative Follow-up Questions |  |
i-MedRAG uses follow-up queries for complex medical questions. |
| Literature QA | Biomedical Literature Q&A System Using Retrieval-Augmented Generation (RAG) |  |
PubMed-centered biomedical literature QA. |
| Medical + Biology QA | MedBioRAG: Semantic Search and Retrieval-Augmented Generation with Large Language Models for Medical and Biological QA |  |
Semantic and lexical search for biomedical QA. |
| Guideline Grounding | Refine Medical Diagnosis Using Generation Augmented Retrieval from Clinical Practice Guidelines |  |
CPG-grounded diagnosis refinement. |
| Orthopedic Chatbot | Development and Evaluation of a Retrieval-Augmented Generation Chatbot for Orthopedic Health Information in German |  |
German evidence-based orthopedic patient information. |
| Agentic Clinical RAG | A Self-Correcting Agentic Graph RAG for Clinical Decision Support in Hepatology |  |
Self-correcting agentic GraphRAG for hepatology. |
| Low-Resource QA | Two-Layer RAG Framework for Low-Resource Medical QA Using Reddit Data |  |
Low-resource medical QA with community data. |
| Nutrigenetics | A Retrieval-Augmented Generation Application for Question-Answering in Nutrigenetics | |
Nutrigenetic GraphRAG application. |
| Multimodal Medical RAG | Iterative Multimodal Retrieval-Augmented Generation for Medical Question Answering |  |
MED-VRAG retrieves PMC page images, tables, figures, and layouts. |
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Rationale-Guided | Rationale-Guided Retrieval Augmented Generation for Medical Question Answering |   |
RAG² filters distractors and uses rationales as retrieval queries. |
| Diagnostic Evaluation | RAG-X: Systematic Diagnosis of Retrieval-Augmented Generation for Medical Question Answering |  |
Disentangles retriever vs generator errors with CUE metrics. |
| Contradiction-aware | Contradictions in Context: Challenges for Retrieval-Augmented Generation in Healthcare |  |
Tests outdated and contradictory evidence in healthcare RAG. |
| RAG Certainty | RAG Certainty: Quantifying the Certainty of Context-Based Responses by LLMs | |
Certainty metric for context-grounded answers. |
| Confidence | Understanding the Impact of Confidence in Retrieval Augmented Generation: A Case Study in the Medical Domain |  |
Studies retrieval confidence and medical answer quality. |
| Adaptive Control | TC-RAG: Turing-Complete RAG's Case Study on Medical LLM Systems | |
Memory stack, retrieval halting, and state-aware RAG control. |
| RAG Credibility | How Credible Is an Answer From Retrieval-Augmented LLMs? | |
Credibility estimation for multi-hop QA. |
| Context Use | Desiderata for the Context Use of Question Answering Systems |   |
How QA systems should use retrieved context. |
| Uncertainty | Modeling Uncertainty and Using Post-fusion as Fallback Improves Retrieval-Augmented Generation | |
Uncertainty-aware fallback for RAG. |
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Omni-RAG | Towards Omni-RAG: Comprehensive Retrieval-Augmented Generation for Large Language Models in Medical Applications | |
Source planning across diverse medical knowledge sources. |
| Adaptive RAG | Adaptive-RAG: Learning to Adapt RAG LLMs through Question Complexity |   |
Chooses retrieval strategy based on question complexity. |
| RAG Relevance | DR-RAG: Applying Dynamic Document Relevance to Retrieval-Augmented Generation |  |
Dynamic document relevance. |
| Rich Queries | RichRAG: Crafting Rich Responses for Multi-faceted Queries in RAG |  |
Better responses for multi-faceted queries. |
| Dynamic Rank | DynRank: Improve Passage Retrieval with Dynamic Zero-Shot Prompting |  |
Dynamic zero-shot prompting for retrieval. |
| Answer Scent | ASRank: Zero-Shot Re-Ranking with Answer Scent for Document Retrieval |  |
Zero-shot re-ranking with answer scent. |
| Passage Expansion | Knowledge Graph-Guided Retrieval Augmented Generation (KG²RAG) |  |
KG-guided chunk expansion. |
| Knowledge Selection | FlexiQA: Leveraging LLM's Evaluation Capabilities for Flexible Knowledge Selection | |
Flexible knowledge selection before generation. |
| Embedding Efficiency | 4-bit Quantization in Vector-Embedding for RAG |  |
Efficient vector embeddings for large retrieval indexes. |
| Low-Resource Retrieval | Unsupervised Domain Adaptation of Dense Retrieval via Zero-Shot Sim Transfer |  |
Useful for low-resource biomedical or multilingual retrieval. |
| Context Generation | Imagination Augmented Generation: Learning to Imagine Richer Context for Question Answering over Large Language Models |  |
Generates richer intermediate context for QA. |
Caution
High-Stakes & Regulatory Compliance
This section covers applications in sensitive domains where error tolerance is low and privacy is paramount. Papers here explore Explainability in financial forecasting, Anonymity in legal court decisions, and Empathy in mental health support. The focus is on domain-specific adaptation and ethical guardrails.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Mental Health QA | MentalQA: An Annotated Arabic Corpus for Questions and Answers of Mental Healthcare |  |
Arabic Mental Healthcare QA. |
| Mental Health | Chinese MentalBERT: Domain-Adaptive Pre-training on Social Media for Chinese Mental Health |  |
Chinese MentalBERT. |
| Mental Privacy | Privacy Aware Question-Answering System for Online Mental Health Risk Assessment | |
Privacy-aware QA. |
| Financial QA | FinReflectKG - MultiHop: Financial QA Benchmark for Reasoning with KG Evidence |  |
Financial multi-hop reasoning. |
| Finance KG | FinQA: A Training-Free Dynamic Knowledge Graph QA System in Finance | |
FinQA (Training-Free). |
| Legal LLM | InLegalLLaMA: Indian Legal Knowledge Enhanced Large Language Model | |
Indian Legal Knowledge LLM. |
| Legal RAG | Interpretable Long-Form Legal QA with Retrieval-Augmented LLMs |   |
Interpretability in Legal QA. |
| Legal GraphRAG | Myanmar Law Cases and Proceedings Retrieval with GraphRAG | |
GraphRAG for Law. |
| Legislative RAG | LexDrafter: Terminology Drafting for Legislative Documents using RAG |  |
Legislative drafting. |
| Legal Retriever | UniLR: Unleashing the Power of LLMs on Multiple Legal Tasks with a Unified Legal Retriever | |
Unified Legal Retriever. |
| Re-Identification | Anonymity at Risk? Assessing Re-Identification Capabilities of LLMs in Court Decisions |  |
Re-ID in court decisions. |
| Crypto QA | CryptOpiQA: A new Opinion and Question Answering dataset on Cryptocurrency |  |
Crypto sentiment/QA. |
| Geopolitical | Scaling LLM-Based Knowledge Graph Generation: A Case Study of Italian Geopolitical News | |
Italian Geopolitical KG. |
| e-Governance | An Open-Domain QA System for e-Governance |   |
Open-domain QA. |
| Education | Enhancing Textbook Question Answering with KG-Augmented LLMs |  |
KG-augmented Textbook QA. |
| Education | Introducing CQuAE: A New French Contextualised QA Corpus for Education | |
CQuAE educational corpus. |
| Grading | SteLLA: A Structured Grading System Using LLMs with RAG |  |
SteLLA grading system. |
| Linguistics | ELQA: A Corpus of Metalinguistic Questions and Answers about English |  |
Metalinguistic QA corpus. |
| Consumer Health | Consumer Health Question Answering Using Off-the-Shelf Components | |
Off-the-shelf components. |
| Health Prompting | Enhancing Consumer Health Question Reformulation: Chain-of-Thought Prompting Integrating Focus, Type, and User Knowledge Level |  |
CoT for question reformulation. |
Note
Evaluating the State-of-the-Art
This section is a comprehensive repository of the gold-standard datasets required to train and evaluate medical QA systems. It spans Multi-Choice Licensing Exams (USMLE, Chinese Medical Exam), Open-Ended Clinical QA, and specialized Shared Tasks (e.g., BioASQ). These resources are essential for benchmarking performance across different modalities, languages, and reasoning complexities.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Medical Dataset | Huatuo-26M: A Large-Scale Chinese Medical Question Answering Dataset |  |
26M QA pairs dataset. |
| Multi-Subject QA | MedMCQA: A Large-scale Multi-Subject Multi-Choice Dataset for Medical domain QA |   |
194k Indian medical questions. |
| Open Domain QA | MedQA: What Disease does this Patient Have? (USMLE-based) |   |
MedQA (USMLE-based). |
| Clinical Benchmark | M-QALM: A Benchmark to Assess Clinical Reading Comprehension and Knowledge Recall |  |
Clinical Reading Comp & Recall. |
| French Benchmark | DrBenchmark: A Large Language Understanding Evaluation Benchmark for French Biomedical Domain | |
20 tasks for French biomedical. |
| Multilingual Med | Towards Building Multilingual Language Model for Medicine (MMedBench) |  |
MMedBench dataset. |
| Strategy QA | StrategyQA: Did Aristotle Use a Laptop? A QA Benchmark with Implicit Reasoning Strategies |   |
Implicit reasoning benchmark. |
| Biomed Dataset | PubMedQA: A Dataset for Biomedical Research Question Answering |   |
PubMedQA dataset. |
| Multilingual FAQ | MFAQ: a Multilingual FAQ Dataset | |
6M FAQ pairs. |
| Literary Fiction | LFED: A Literary Fiction Evaluation Dataset for Large Language Models |  |
Literary fiction eval. |
| Dynamic QA | Let LLMs Take on the Latest Challenges! A Chinese Dynamic QA Benchmark | |
Dynamic QA benchmark. |
| Arabic Dataset | ArabicaQA: A Comprehensive Dataset for Arabic Question Answering |  |
Comprehensive Arabic QA. |
| Chinese Benchmark | CMB: A Comprehensive Medical Benchmark in Chinese | |
Chinese medical benchmark. |
| French Medical | FrenchMedMCQA: A French Multiple-Choice QA Dataset for Medical Domain |  |
French Multi-Choice Medical QA. |
| Realistic QA | RealMedQA: A Pilot Biomedical QA Dataset Containing Realistic Clinical Questions |  |
Real clinical questions (NICE). |
| Pharmacist Exam | ExplainCPE: A Free-text Explanation Benchmark of Chinese Pharmacist Examination | |
Pharmacist exam explanations. |
| Arabic Health | AraHealthQA 2025: The First Shared Task on Arabic Health QA |   |
Shared task (Mental/General). |
| Japanese Eval | JMedBench: A Benchmark for Evaluating Japanese Biomedical LLMs | |
JMedBench. |
| EHR QA | DrugEHRQA: A QA Dataset on Structured and Unstructured EHR | |
DrugEHRQA. |
| Radiology QA | RadQA: A QA Dataset to Improve Comprehension of Radiology Reports | |
RadQA. |
| Clinical QA | RJUA-QA: A Comprehensive QA Dataset for Clinical Reasoning in Urology |  |
Urology clinical reasoning. |
| Short-Answer | ACTA: Short-Answer Grading in High-Stakes Medical Exams | |
Grading medical exams. |
| Arabic Wikipedia | ArTrivia: Harvesting Arabic Wikipedia to Build A New Arabic QA Dataset | |
Arabic Wikipedia QA dataset. |
| Scientific QA | SciQA: Extensive Analysis of the SciQA Benchmark with LLMs |  |
SciQA benchmark analysis. |
Important
Beyond Text: The Convergence of Vision and Language
This section explores the frontier of Multimodal AI, where Large Language Models (LLMs) connect with visual data to "see" and interpret medical contexts. It covers Medical Visual Question Answering (Med-VQA), Radiology Report Generation, and Multimodal Knowledge Graphs. These resources are pivotal for systems that must reason over heterogeneous data sources, such as aligning clinical notes with pixel-level evidence from X-rays, CT scans, and pathology slides.
| Topic | Full Title | Resources | Notes |
|---|---|---|---|
| Multimodal QA | CLIPSyntel: CLIP and LLM Synergy for Multimodal Question Summarization in Healthcare |  |
CLIP + LLM for summarization. |
| Multimodal KG | VisualSem: A High-Quality Knowledge Graph for Vision and Language |  |
VisualSem KG. |
| Visual QA | Seeing Beyond: Enhancing Visual Question Answering with Multi-Modal Retrieval | |
Multi-modal retrieval for VQA. |
| EHR + CXR | EHRXQA: A Multi-Modal QA Dataset for Electronic Health Records with Chest X-ray Images |   |
Multi-modal EHR QA. |
| VQA Integration | Modality-Aware Integration with LLMs for Knowledge-Based Visual Question Answering |   |
MAIL framework for VQA. |
| Multimodal Sentiment | ConKI: Contrastive Knowledge Injection for Multimodal Sentiment Analysis |  |
Contrastive knowledge injection. |
| Visual Reasoning | Med-R1: Reinforcement Learning for Generalizable Medical Reasoning in VLMs |  |
RL for VLMs in medicine. |
| Vietnamese VQA | Enhancing Vietnamese VQA through Curriculum Learning on Raw and Augmented Text |  |
Vietnamese VQA. |
| Drug Interaction | MKG-FENN: A Multimodal KG Fused End-to-End Neural Network for DDI Prediction |  |
Drug-drug interaction. |
| Negative Sampling | Modality-Aware Negative Sampling for Multi-modal Knowledge Graph Embedding | |
MANS negative sampling. |
| Aspect-aware KG | AspectMMKG: A Multi-modal Knowledge Graph with Aspect-aware Entities |  |
AspectMMKG. |
| Fact Verification | Multi-source Knowledge Enhanced Graph Attention Networks for Multimodal Fact Verification | |
Multi-source graph attention. |
| Entity Tagging | Multimodal Entity Tagging with Multimodal Knowledge Base | |
Multimodal Entity Tagging (MET). |
| Noise-powered | Noise-powered Multi-modal Knowledge Graph Representation Framework | |
SNAG method. |
| Spatial Expl | Spatially Grounded Explanations in Vision-Language Models for Document VQA |  |
Spatially grounded explanations. |
| Structure Guided | SGMEA: Structure-Guided Multimodal Entity Alignment | |
SGMEA. |
| Spoken MRC | VlogQA: Task, Dataset, and Baseline Models for Vietnamese Spoken-Based MRC | |
VlogQA. |
| Multimodal Doc | T2KG: Transforming Multimodal Document to Knowledge Graph |  |
T2KG. |
| Visual Rel | Visual Relationship Detection With Visual-Linguistic Knowledge | |
Visual relationship detection. |
| Zero-Shot Rel | Zero-Shot Relational Learning for Multimodal Knowledge Graphs |  |
Zero-shot relational learning. |
| Multimodal Repr | Enhancing Multimodal Knowledge Graph Representation Learning through Triple Contrastive Learning | |
Triple contrastive learning. |
This repository chronicles a pivotal shift in Medical AI from static Information Retrieval to dynamic Clinical Reasoning. As the field matures, three defining paradigms are emerging:
- The Agentic Shift: Moving beyond passive chatbots to proactive Copilots. Systems like MedAgents and Dr. Copilot demonstrate that the future lies in LLMs that can plan, self-correct, and execute multi-step diagnostic workflows.
- Multimodal Synergy: True clinical understanding requires seeing as well as reading. The next generation of models achieves Multimodal Fluency, seamlessly synthesizing pixel-level evidence (X-rays, Pathology) with textual knowledge (Guidelines, EHRs).
- Democratization & Safety: As capabilities scale, so must responsibility. The focus is pivoting toward Privacy-Preserving RAG and Linguistic Equity, ensuring that life-saving AI is robust, compliant, and accessible across all languages from English to Amharic.
Tip
Help this list grow
Contributions are welcome, especially for 2025-2026 medical RAG, GraphRAG, multilingual clinical QA, medical safety evaluation, and open-source implementations.
| Contribution Type | What to Add | Preferred Evidence |
|---|---|---|
| Paper | Title, venue/year, arXiv/ACL/PubMed/DOI link | Peer-reviewed venue or stable preprint. |
| Code | Official GitHub, HuggingFace, dataset, project page | Maintained repository or reproducible artifact. |
| Benchmark | Dataset, task, metrics, leaderboard | Clear evaluation protocol and license. |
| Medical Safety | Guardrails, contradiction checks, citation/grounding metrics | Human/clinical evaluation when available. |
| Multilingual | Non-English datasets, models, or RAG pipelines | Language coverage and cultural/clinical relevance. |
Note
If this repository helps your research, please consider giving it a ⭐ so more medical AI researchers can find it.
This repository is licensed under the Creative Commons Attribution 4.0 International License, granting you the freedom to share and adapt this work for any purpose even commercially as long as you provide appropriate credit to the original research paper authors.
