[Integration of large language models into the clinic : Revolution in analysing and processing patient data to increase efficiency and quality in radiology].

Abstract

Background: Large Language Models (LLMs) like ChatGPT, Llama and Claude are transforming healthcare by interpreting complex text, extracting information, and providing guideline-based support. Radiology, with its high patient volume and digital workflows, is a ideal field for LLM integration.

Objective: Assessment of the potential of LLMs to enhance efficiency, standardization, and decision support in radiology, while addressing ethical and regulatory challenges.

Material and methods: Pilot studies at Freiburg and Basel university hospitals evaluated local LLM systems for tasks like prior report summarization and guideline-driven reporting. Integration with Picture Archiving and Communication System (PACS) and Electronic Health Record (EHR) systems was achieved via Digital Imaging and Communications in Medicine (DICOM) and Fast Healthcare Interoperability Resources (FHIR) standards. Metrics included time savings, compliance with the European Union (EU) Artificial Intelligence (AI) Act, and user acceptance.

Results: LLMs demonstrate significant potential as a support tool for radiologists in clinical practice by reducing reporting times, automating routine tasks, and ensuring consistent, high-quality results. They also support interdisciplinary workflows (e.g., tumor boards) and meet data protection requirements when locally implemented.

Discussion: Local LLM systems are feasible and beneficial in radiology, enhancing efficiency and diagnostic quality. Future work should refine transparency, expand applications, and ensure LLMs complement medical expertise while adhering to ethical and legal standards.