Retrieval-augmented generation (RAG) is an emerging technique that enhances large language models (LLMs) by enabling them to access and incorporate external knowledge sources during response generation. In radiology, in which clinical accuracy, guideline adherence, and contextual understanding are critical, RAG offers a promising approach for supporting decision-making, reporting, and patient communication. Unlike stand-alone LLMs, which rely solely on pretraining, RAG models retrieve relevant data, such as imaging guidelines, prior reports, or literature, to generate relevant outputs. This approach bridges the gap between the static knowledge of traditional models and the dynamic nature of clinical radiology, helping to reduce the risk of outdated or inaccurate information influencing decisions. Understanding RAG-enabled technology allows radiologists to better evaluate artificial intelligence tools, advocate for safe deployment, and engage in innovation. This article introduces RAG in practical terms, emphasizing what radiologists need to know to apply or assess its use in everyday practice.
Dietrich et al. (2026) studied this question.