Infherno: End-to-end Agent-based FHIR Resource Synthesis from Free-form Clinical Notes
自由記述の診療記録からFHIRリソースを合成するエージェント型アプローチ
A new method was developed for the end-to-end, agent-based synthesis of FHIR resources directly from free-form clinical notes.
Summary
This paper proposes 'Infherno,' an end-to-end framework that utilizes LLM agents and external tools to synthesize FHIR resources from unstructured clinical text. This approach addresses the limitations of conventional systems regarding generalizability and structural conformity.
Details
This paper introduces 'Infherno,' a novel, agentic approach designed to automatically generate structured FHIR resources from unstructured clinical notes, adhering to the HL7 FHIR standard. Traditional information extraction (IE) methods or simple LLM translations often struggle with complex, variable clinical contexts due to their reliance on fixed schemas. The proposed Infherno framework overcomes these limitations by employing an agentic LLM approach. It utilizes external tools—such as SNOMED CT terminology databases and code execution environments—to perform multi-step reasoning processes. This ensures not only accurate translation but also high semantic correctness and structural conformity to the FHIR schema. This capability is crucial for clinical data integration, enabling healthcare providers to standardize and retrieve data from diverse sources into a reusable format (FHIR/JSON). It holds significant relevance for advancing electronic health records and building large-scale data analysis infrastructure in Japanese healthcare settings.
Technology Note
FHIR（Fast Healthcare Interoperability Resources）は医療データ交換の国際標準。このエントリの関連技術: FHIR
Original content copyright by respective publishers