Large language models (LLMs) in radiography research: A narrative review

Artificial intelligence (AI) has become increasingly embedded in Radiography research and practice, extending beyond diagnostic support and workflow optimisation to non-patient-facing applications. Generative AI (GenAI), particularly Large Language Models (LLMs), have been used in radiography research, generating synthetic data, assisting in literature reviews, and facilitating multilingual communication. This narrative review aims to explore the opportunities, challenges, and implications of LLM integration in radiography research.

A narrative review approach was employed, identifying relevant publications and integrating author-led examples, including survey design, translation support, and workflow integration. The literature shows that LLMs accelerate research processes by supporting systematic literature retrieval, enhancing survey development, generating synthetic imaging data, streamlining data analysis and enhancing internationalisation. Case studies highlighted measurable benefits, such as improved CT image quality and reduced examination times through AI-assisted communication. However, challenges included hallucinated outputs, embedded biases, risks to privacy, regulatory challenges and environmental costs of model training.

GenAI and LLMs offer transformative opportunities for radiography research across multiple stages, from study design to dissemination. Nonetheless, integration must be accompanied by validation against expert-reviewed datasets, transparent reporting, and ethical safeguards to ensure reliability.

Radiography researchers should adopt GenAI tools with a "trust but verify" approach: leveraging their efficiency while verifying outputs through expert oversight. Training, governance, and validation frameworks are essential for safe implementation, ensuring these technologies augment rather than replace human expertise in radiography research.