Agentic systems in radiology: Principles, opportunities, privacy risks, regulation, and sustainability concerns

Area of Science:

• Artificial Intelligence in Radiology
• Medical Imaging Informatics
• Computational Pathology

Background:

• Transformer-based large language models (LLMs) show promise for automating radiology tasks like report generation.
• Conventional LLMs are limited by their inability to interact with external systems, hindering clinical workflow integration.
• Agentic systems offer a new paradigm by embedding LLMs within frameworks enabling reasoning, planning, and action.

Purpose of the Study:

• To provide a comprehensive overview of agentic systems in medical imaging and radiology.
• To summarize key developments and applications of agentic systems in radiology.
• To discuss the potential benefits and challenges of integrating agentic systems into clinical practice.

Main Methods:

• Review of existing literature on agentic systems and LLMs in medical imaging.
• Analysis of agentic system architectures and operational mechanisms.
• Examination of multi-agent frameworks for automated radiomics pipelines.

Main Results:

• Agentic systems extend LLM capabilities for dynamic interaction with users, tools, and data sources.
• Recent multi-agent frameworks show potential for automated radiomics pipelines.
• These systems can enhance reproducibility, interpretability, and accessibility of AI-driven radiology workflows.

Conclusions:

• Agentic systems represent a significant advancement for AI in radiology, enabling more sophisticated automation and decision support.
• Addressing regulatory, ethical, and sustainability challenges is crucial for safe and responsible clinical integration.
• Further research is needed to bridge existing gaps for widespread adoption of agentic AI in medical imaging.