Large Language Models in the Radiology Workflow: Current Applications, Challenges, and Future Perspectives.
Authors
Abstract
Large language models (LLMs) have attracted increasing research interest in radiology owing to their capacity for natural language processing and, more recently, multimodal image-text reasoning. This review provides a clinically oriented overview of LLM applications across the radiology workflow, encompassing scan protocols, image interpretation, radiology reporting, and communication with patients and clinicians. Although the existing general-purpose multimodal LLMs show promise in diagnostic tasks, their image-interpretation performance is limited; they often prioritize textual context over visual information, and they are prone to hallucinations. In contrast, LLMs have demonstrated practical utility in radiology reporting, including structured radiology reporting, summarization, automated data extraction, and clinically meaningful report categorization. This review also discusses the reporting guidelines and performance evaluation methodologies specific to LLM research to promote more scientifically rigorous and reproducible studies in this rapidly evolving field. Finally, we outline the future directions for LLM research to support the responsible and effective integration of LLMs in radiology practice.