Country / Region
EMEA
Tags
Artificial intelligence, Clinical Practice, Data quality, Implementation, Innovation
In this work, we present the technical architecture behind an AI-powered application that automatically extracts diagnoses from clinical documentation and updates patient problem lists using SNOMED CT. We explore how large language models combined with semantic search capabilities enable accurate identification and coding of diagnoses, while maintaining clinical relevance and addressing the technical challenges of integrating with existing EHR systems in a Belgian hospital.
Description
This project focuses on the development of a hybrid AI system combining large language models and semantic search for clinical text analysis. We created a pipeline to identify and extract new diagnoses from unstructured clinical documentation as well as mapping extracted diagnoses to appropriate SNOMED CT concepts. The scope includes development of a validation framework to ensure accuracy before updating problem lists and integration with existing EHR systems at a hospital setting with focus on minimal disruption of clinical workflows.
Scope
SNOMED CT's multilingual capabilities support deployment across linguistically diverse healthcare settings, and the standard's widespread adoption in healthcare IT systems facilitates integration with existing EHR infrastructures. Moreover, its granularity allows for highly specific encoding of clinical concepts, enabling a precise representation of clinical diagnoses.
How SNOMED CT will be used
SNOMED CT serves as the foundational clinical terminology for structuring and validating extracted diagnoses from unstructured clinical text. SNOMED CT is used as a target ontology for mapping identified conditions and as a knowledge base to support semantic search.
Why SNOMED CT will be used
Contact
