Large Language Models for Structured Reporting in Radiology

This narrative review examines the application of large language models (LLMs) in structured reporting (SR) within the field of radiology. It discusses the historical context and evolution of SR, highlighting its aim to standardize and enhance the quality of radiology reports. Despite evidence supporting the benefits of SR, such as reducing errors and improving adherence to guidelines, its adoption has been limited. The review identifies ten studies focusing on LLMs, particularly generative pre-trained transformers (GPT)-3.5 and GPT-4, which show promising results, including the feasibility of multilingual applications. The authors outline the limitations and regulatory challenges faced by LLMs in clinical practice, emphasizing the need for overcoming issues related to algorithm transparency and training data. The review concludes by underscoring the potential of LLMs to transform radiology report processing and facilitate the broader implementation of SR, provided that existing challenges are addressed.