HealthRecordCommunity
FHIRLOINCSNOMED CT🌏 Germany🏛 Fraunhofer-GesellschaftEnriched

AI-Powered Structuring of Medical Data: Overview of the FHIR-Starter Project

Fraunhofer research project “FHIR-Starter” develops AI-based software for structuring medical

Fraunhofer-Gesellschaft developed AI-based software designed to structure medical data using FHIR standards.

February 21, 2025Fraunhofer-Gesellschaft

Summary

Fraunhofer Institute launched the 'FHIR-Starter' project to automatically extract and standardize data from free-text medical documents (like PDFs) using Large Language Models (LLMs) and Natural Language Processing (NLP). This aims to solve key challenges in German healthcare data utilization.

Key Players

Fraunhofer IESE

Details

The project addresses critical issues within Germany’s lagging digitalization of the healthcare system. Currently, clinical findings are often communicated via multi-page PDF documents, requiring time-consuming manual reading and transfer into practice management or hospital information systems (HIS), which is prone to errors. Furthermore, utilizing data for medical research from full texts is difficult. 'FHIR-Starter' aims to develop a software service that uses LLMs and NLP to analyze these free-text documents and convert them into standardized formats, specifically leveraging the FHIR standard alongside LOINC and SNOMED-CT coding systems. This enhances data interoperability for healthcare data exchange. The project tackles two major technical challenges: ensuring data reliability (preventing AI 'hallucinations') and maintaining GDPR compliance. For security, it plans to use open-source LLMs running on users' own servers, creating a self-contained and secure system. The potential applications include advancing electronic patient records (ePA) by allowing the display of longitudinal lab values or automatically generating medication lists.

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

📰
Read Original Article
iese.fraunhofer.de
→

Original content copyright by respective publishers