Agentic AI and Large Language Models in Radiology: Opportunities and Hallucination Challenges

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology Workflow Optimization
• Natural Language Processing in Healthcare

Background:

• Large language models (LLMs) are increasingly used in radiology.
• Hallucinations (generating incorrect information) are a major trust barrier for LLMs in this field.

Purpose of the Study:

• To review agentic artificial intelligence (AI) approaches for reducing LLM hallucinations in radiology.
• To assess the potential of multi-agent systems, retrieval-augmented generation (RAG), and uncertainty quantification.

Main Methods:

• Comprehensive review of emerging agentic AI strategies.
• Evaluation of multi-agent frameworks, RAG, and uncertainty quantification.
• Analysis of evidence from 2024-2025 on diagnostic accuracy and error reduction.

Main Results:

• Agentic AI can improve diagnostic accuracy and reduce errors in radiology.
• Multi-agent systems offer cross-validation and workflow orchestration.
• RAG enhances accuracy by grounding responses in medical literature.
• Uncertainty quantification aids collaborative analysis within multi-agent systems.

Conclusions:

• Agentic AI, particularly multi-agent frameworks and RAG, shows significant promise for mitigating LLM hallucinations in radiology.
• Computational demands and lack of comprehensive clinical validation are current limitations.
• Safe clinical deployment requires human oversight, tailored evaluation metrics, and regulatory adaptation.