Trends and Trajectories in the Rise of Large Language Models in Radiology: Scoping Review.

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Informatics
• Radiology AI Applications

Background:

• Large language models (LLMs) are rapidly advancing, with increasing applications in radiology, including report generation, decision support, and workflow optimization.
• Despite their potential, a comprehensive evaluation of LLM applications, performance, and limitations within the radiology domain is currently lacking.

Purpose of the Study:

• To systematically map the current applications of LLMs in radiology.
• To evaluate the performance of LLMs across key radiological tasks.
• To identify limitations and future research directions for LLMs in radiology.

Main Methods:

• A scoping review was conducted following the Arksey and O'Malley framework and PRISMA-ScR guidelines.
• Searched PubMed, ScopusCOPUS, and IEEE Xplore for peer-reviewed studies from January 2022 to December 2024.
• Included empirical evaluations of LLMs in radiology; excluded commentaries and technical proposals without evaluation.

Main Results:

• 67 studies were included, with GPT-4 being the most common LLM (42%).
• Primary use cases included decision support (58%), report generation/summarization (24%), and workflow optimization (18%).
• LLMs excelled in structured text tasks (>94% accuracy) but showed variable diagnostic performance (16%-86%) due to dataset bias and limited validation. Most studies were single-center proof-of-concept (79.1%).

Conclusions:

• LLMs demonstrate significant potential to enhance radiological workflows, especially for structured reporting and summarization.
• Current diagnostic performance is inconsistent, and robust external validation is needed for clinical integration.
• Future research should focus on prospective, multicenter validation of domain-adapted and multimodal LLMs for safe clinical use.