MedQGraph: TMKGs for AI-Driven Healthcare Insights
summary
Prototype converts MIMIC‑IV (~300k ICU encounters) into a temporal medical knowledge graph on Neo4j Aura, using Neo4j GDS for embeddings and GPT-4 entity extraction to pull diagnoses, meds, labs and timestamps. Cypher queries run in milliseconds to reconstruct patient journeys, an LLM generates Cypher for natural‑language questions, and connectors (Kafka CDC, SNOMED CT, RxNorm) plus probabilistic patient matching support multi‑vendor ingestion and near‑real‑time updates. Automated voice agents and zero‑downtime claims round out the stack.
AI-written summary · about
chapters (16)
- 0:00Introduction and team
- 0:25Problem with current EHR systems
- 1:50Why conventional EHR struggles
- 2:34Data sources for knowledge graphs
- 3:17Building temporal medical knowledge graph
- 4:19Extracting entities with GPT-4
- 5:01Querying the knowledge graph
- 6:24RAG plus knowledge graph approach
- 7:25Automated calling and summarization
- 8:32Medical transcript ontology
- 9:55Agentic model for natural queries
- 11:41System architecture overview
- 12:26Security and compliance focus
- 13:52Scaling to multi-vendor EHRs
- 15:36Challenges and future work
- 15:58Q&A and closing
related talks
32mGraph-based ML for Automated Health Care Services
28mRole of Knowledge Graphs in Mental Health Diagnosis and Cure
59mArtificial intelligence in Healthcare
15mLightning Talk: Future-Proofing Compliance: Leveraging Knowledge Graphs and AI in Cybe...
15mEnhancing Medical Data Processing With Neo4j
05mHow generative AI can support medical staff
25mAdaptive GraphRAG: A Framework for Knowledge Graph Quality, Consistency, Evolution
01h 26mThinking in Graphs: The Cognitive Foundations of Agentic AI